The stochastic gradient descent (SGD) optimizers are generally used to train the convolutional neural networks (CNNs). In recent years, several adaptive momentum based SGD optimizers have been introduced, such as Adam, diffGrad, Radam and AdaBelief. However, the existing SGD optimizers do not exploit the gradient norm of past iterations and lead to poor convergence and performance. In this paper, we propose a novel AdaNorm based SGD optimizers by correcting the norm of gradient in each iteration based on the adaptive training history of gradient norm. By doing so, the proposed optimizers are able to maintain high and representive gradient throughout the training and solves the low and atypical gradient problems. The proposed concept is generic and can be used with any existing SGD optimizer. We show the efficacy of the proposed AdaNorm with four state-of-the-art optimizers, including Adam, diffGrad, Radam and AdaBelief. We depict the performance improvement due to the proposed optimizers using three CNN models, including VGG16, ResNet18 and ResNet50, on three benchmark object recognition datasets, including CIFAR10, CIFAR100 and TinyImageNet.

As more and more office documents are captured, stored, and shared in digital format, and as image editing software are becoming increasingly more powerful, there is a growing concern about document authenticity. To prevent illicit activities, this paper presents a new method for detecting altered text in document images. The proposed method explores the relationship between positive and negative coefficients of DCT to extract the effect of distortions caused by tampering by fusing reconstructed images of respective positive and negative coefficients, which results in Positive-Negative DCT coefficients Fusion (PNDF). To take advantage of spatial information, we propose to fuse R, G, and B color channels of input images, which results in RGBF (RGB Fusion). Next, the same fusion operation is used for fusing PNDF and RGBF, which results in a fused image for the original input one. We compute a histogram to extract features from the fused image, which results in a feature vector. The feature vector is then fed to a deep neural network for classifying altered text images. The proposed method is tested on our own dataset and the standard datasets from the ICPR 2018 Fraud Contest, Altered Handwriting (AH), and faked IMEI number images. The results show that the proposed method is effective and the proposed method outperforms the existing methods irrespective of image type.

Appropriate modeling of a surveillance scene is essential for the detection of anomalies in road traffic. Learning usual paths can provide valuable insight into road traffic conditions and thus can help in identifying unusual routes taken by commuters/vehicles. If usual traffic paths are learned in a nonparametric way, manual interventions in road marking can be avoided. In this paper, we propose an unsupervised and nonparametric method to learn the frequently used paths from the tracks of moving objects in Θ(kn) time, where k denotes the number of paths and n represents the number of tracks. In the proposed method, temporal dependencies of the moving objects are considered to make the clustering meaningful using temporally incremental gravity model (TIGM). In addition, the distance-based scene learning makes it intuitive to estimate the model parameters. Further, we have extended the TIGM hierarchically as a dynamically evolving model (DEM) to represent notable traffic dynamics of a scene. The experimental validation reveals that the proposed method can learn a scene quickly without prior knowledge about the number of paths ( k ). We have compared the results with various state-of-the-art methods. We have also highlighted the advantages of the proposed method over the existing techniques popularly used for designing traffic monitoring applications. It can be used for administrative decision making to control traffic at junctions or crowded places and generate alarm signals, if necessary.

Writer verification has drawn significant attention over the past few decades due to its extensive applications in forensics and biometrics. In traditional writer verification, handwriting similarity/dissimilarity analysis is mostly performed by extracting two feature vectors from two respective handwritten samples, followed by comparing them in relation to their similarity. In the state-of-the-art writer verification approaches, a distance metric is usually employed in terms of the similarity between two handwritten samples. If the distance between two handwritten samples is greater than a given threshold, then the samples are assumed to be written by two different writers, otherwise, they are considered to be due to the same writer. In this paper, for the very first time, we propose a model that generates English sentences to explain reasons for writer dissimilarity/similarity. First, our proposed model obtains features from handwritten images by employing a convolutional neural network, verifies the writer using a Siamese architecture, and generates English words using a recurrent neural network. Finally, these two networks are merged using an affine transformation to produce an explanatory sentence in support of writer similarity/dissimilarity. We evaluated our model on a handwritten numeral database of 100 writers and obtained promising results.

In this study, a robust algorithm for dewarping of camera‐captured document images, mainly in Bangla script, is proposed. The algorithm can handle various types of warped document images and they are generated due to different types of document surfaces (convex, concave or multi‐folded). The proposed algorithm is independent of font type, font size, font style and camera view angle. After initial preprocessing, the method first demarcates the text lines present in the document image. Then, the headline ( shirorekha ) position of each text line is estimated. Based on the headline position and shape, each text line is dewarped. If the document is highly warped, distorted text (e.g. thinner and shorter characters) is generated after dewarping. Special care has been taken to minimise this distortion based on most undistorted character information. Exhaustive testing shows the robustness and shape improvement of the proposed algorithm. Finally, for shape quality evaluation, some new measures are defined.

Deep learning‐based approaches have become very prominent in recent years due to its outstanding performance as compared to the hand‐extracted feature‐based methods. Convolutional neural network (CNN) is a type of deep learning architecture to deal with the image/video data. Residual network and squeeze and excitation network (SENet) are among recent developments in CNN for image classification. However, the performance of SENet depends on the squeeze operation done by global pooling, which sometimes may lead to poor performance. In this study, the authors propose a bilinear fusion mechanism over different types of squeeze operation such as global pooling and max pooling. The excitation operation is performed using the fused output of squeeze operation. They used to model the proposed fused SENet with the residual unit and name it as FuSENet . Here the classification experiments are performed over benchmark hyperspectral image datasets. The experimental results confirm the superiority of the proposed FuSENet method with respect to the state‐of‐the‐art methods. The source code of the complete system is made publicly available at https://github.com/swalpa/FuSENet .

In this letter, a semisupervised neural network-based approach has been proposed for handling the class-imbalance problem in land-cover classification under a hybrid integration of selective undersampling, oversampling, and a bagging-based ensemble of classifiers. Here, a selective undersampling technique is utilized so as to minimize the loss of information from the majority classes; whereas, the minority class sizes are simultaneously increased by exploiting their presence in the unlabeled test samples. Finally, the imbalanced original training set along with the newly found minority samples is used to classify the remaining unlabeled samples from the test set. Experiments conducted on the patterns collected from multispectral (high as well as very high resolution images) and hyperspectral remote sensing satellite images show encouraging performance of the proposed scheme when compared to other state-of-the-art techniques.

Hyperspectral image (HSI) classification is widely used for the analysis of remotely sensed images. Hyperspectral imagery includes varying bands of images. Convolutional neural network (CNN) is one of the most frequently used deep learning-based methods for visual data processing. The use of CNN for HSI classification is also visible in recent works. These approaches are mostly based on 2-D CNN. On the other hand, the HSI classification performance is highly dependent on both spatial and spectral information. Very few methods have used the 3-D-CNN because of increased computational complexity. This letter proposes a hybrid spectral CNN (HybridSN) for HSI classification. In general, the HybridSN is a spectral-spatial 3-D-CNN followed by spatial 2-D-CNN. The 3-D-CNN facilitates the joint spatial-spectral feature representation from a stack of spectral bands. The 2-D-CNN on top of the 3-D-CNN further learns more abstract-level spatial representation. Moreover, the use of hybrid CNNs reduces the complexity of the model compared to the use of 3-D-CNN alone. To test the performance of this hybrid approach, very rigorous HSI classification experiments are performed over Indian Pines, University of Pavia, and Salinas Scene remote sensing data sets. The results are compared with the state-of-the-art hand-crafted as well as end-to-end deep learning-based methods. A very satisfactory performance is obtained using the proposed HybridSN for HSI classification. The source code can be found at https://github.com/gokriznastic/HybridSN.

Of late, convolutional neural networks (CNNs) find great attention in hyperspectral image (HSI) classification since deep CNNs exhibit commendable performance for computer vision-related areas. CNNs have already proved to be very effective feature extractors, especially for the classification of large data sets composed of 2-D images. However, due to the existence of noisy or correlated spectral bands in the spectral domain and nonuniform pixels in the spatial neighborhood, HSI classification results are often degraded and unacceptable. However, the elementary CNN models often find intrinsic representation of pattern directly when employed to explore the HSI in the spectral-spatial domain. In this article, we design an end-to-end spectral-spatial squeeze-and-excitation (SE) residual bag-of-feature (S3EResBoF) learning framework for HSI classification that takes as input raw 3-D image cubes without engineering and builds a codebook representation of transform feature by motivating the feature maps facilitating classification by suppressing useless feature maps based on patterns present in the feature maps. To boost the classification performance and learn the joint spatial-spectral features, every residual block is connected to every other 3-D convolutional layer through an identity mapping followed by an SE block, thereby facilitating the rich gradients through backpropagation. Additionally, we introduce batch normalization on every convolutional layer (ConvBN) to regularize the convergence of the network and scale invariant BoF quantization for the measure of classification. The experiments conducted using three well-known HSI data sets and compared with the state-of-the-art classification methods reveal that S3EResBoF provides competitive performance in terms of both classification and computation time.

In this paper, we work on intra-variable handwriting, where the writing samples of an individual can vary significantly. Such within-writer variation throws a challenge for automatic writer inspection, where the state-of-the-art methods do not perform well. To deal with intra-variability, we analyze the idiosyncrasy in individual handwriting. We identify/verify the writer from highly idiosyncratic text-patches. Such patches are detected using a deep recurrent reinforcement learning-based architecture. An idiosyncratic score is assigned to every patch, which is predicted by employing deep regression analysis. For writer identification, we propose a deep neural architecture, which makes the final decision by the idiosyncratic score-induced weighted average of patch-based decisions. For writer verification, we propose two algorithms for patch-fed deep feature aggregation, which assist in authentication using a triplet network. The experiments were performed on two databases, where we obtained encouraging results.

Stochastic gradient descent (SGD) is one of the core techniques behind the success of deep neural networks. The gradient provides information on the direction in which a function has the steepest rate of change. The main problem with basic SGD is to change by equal-sized steps for all parameters, irrespective of the gradient behavior. Hence, an efficient way of deep network optimization is to have adaptive step sizes for each parameter. Recently, several attempts have been made to improve gradient descent methods such as AdaGrad, AdaDelta, RMSProp, and adaptive moment estimation (Adam). These methods rely on the square roots of exponential moving averages of squared past gradients. Thus, these methods do not take advantage of local change in gradients. In this article, a novel optimizer is proposed based on the difference between the present and the immediate past gradient (i.e., diffGrad). In the proposed diffGrad optimization technique, the step size is adjusted for each parameter in such a way that it should have a larger step size for faster gradient changing parameters and a lower step size for lower gradient changing parameters. The convergence analysis is done using the regret bound approach of the online learning framework. In this article, thorough analysis is made over three synthetic complex nonconvex functions. The image categorization experiments are also conducted over the CIFAR10 and CIFAR100 data sets to observe the performance of diffGrad with respect to the state-of-the-art optimizers such as SGDM, AdaGrad, AdaDelta, RMSProp, AMSGrad, and Adam. The residual unit (ResNet)-based convolutional neural network (CNN) architecture is used in the experiments. The experiments show that diffGrad outperforms other optimizers. Also, we show that diffGrad performs uniformly well for training CNN using different activation functions. The source code is made publicly available at https://github.com/shivram1987/diffGrad.

In this paper, we undertake a task to find named entities directly from unstructured handwritten document images without any intermediate text/character recognition. Here, we do not receive any assistance from natural language processing. Therefore, it becomes more challenging to detect the named entities. We work on Bengali script which brings some additional hurdles due to its own unique script characteristics. Here, we propose a new deep neural network-based architecture to extract the latent features from a text image. The embedding is then fed to a BLSTM (Bidirectional Long Short-Term Memory) layer. After that, the attention mechanism is adapted to an approach for named entity detection. We perform experimentation on two publicly-available offline handwriting repositories containing 420 Bengali handwritten pages in total. The experimental outcome of our system is quite impressive as it attains 95.43% balanced accuracy on overall named entity detection.

Automated handwriting analysis is a popular area of research owing to the variation of writing patterns. In this research area, writer verification is one of the most challenging branches, having direct impact on biometrics and forensics. In this paper, we deal with offline writer verification on complex handwriting patterns. Therefore, we choose a relatively complex script, i.e., Indic Abugida script Bengali (or, Bangla) containing more than 250 compound characters. From a handwritten sample, the probability distribution functions (PDFs) of some handcrafted features are obtained and input to a convolutional neural network (CNN). For such a CNN architecture, we coin the term "PDFCNN", where handcrafted feature PDFs are hybridized with auto-derived CNN features. Such hybrid features are then fed into a Siamese neural network for writer verification. The experiments are performed on a Bengali offline handwritten dataset of 100 writers. Our system achieves encouraging results, which sometimes exceed the results of state-of-the-art techniques on writer verification.

In this paper, we study handwriting idiosyncrasy in terms of its structural eccentricity. In this study, our approach is to find idiosyncratic handwritten text components and model the idiosyncrasy analysis task as a machine learning problem supervised by human cognition. We employ the Inception network for this purpose. The experiments are performed on two publicly available databases and an in-house database of Bengali offline handwritten samples. On these samples, subjective opinion scores of handwriting idiosyncrasy are collected from handwriting experts. We have analyzed the handwriting idiosyncrasy on this corpus which comprises the perceptive ground-truth opinion. We also investigate the effect of idiosyncratic text on writer identification by using the SqueezeNet. The performance of our system is promising.

In AEC (Architecture, Engineering & Construction) industry, drawing documents are used as a blueprint to facilitate the construction process. It is also represented as a graphical language that communicates ideas and information from one mind to another. In AEC documents, text is present in Machine-printed and hand-written format. Since the algorithms for recognition of machine-printed and hand-written texts are different, it is important to distinguish between these two types of texts before sending the document to respective recognition system. In this paper we proposed a novel approach for the classification machine-printed and hand-written text from AEC Documents. Before Classification Hand-Written and Machine-Printed text from the documents our system used some preprocessing which includes binarization, text graphics separation and word segmentation. The Words are segmented based on certain structural properties of Isothetic Covers (IC) tightly enclosing the words in a document. The grid size properties of IC are selected by some statistical analysis of connected component of the document. Then Word level Gabor Filter based features are extracted with spooling information for classification. A standard classifier based on SVM is used to classify the text. This task is performed at word level of AEC documents and we achieved an overall accuracy of 98.45%.

In this paper, we study the difficulties arising in reading and writing of Bengali conjunct characters by human-beings. Such difficulties appear when the human cognitive system faces certain obstructions in effortlessly reading/writing. In our computer-based investigation, we consider the reading/writing difficulty analysis task as a machine learning problem supervised by human perception. To this end, we employ two distinct models: (a) an auto-derived feature-based Inception network and (b) a hand-crafted feature-based SVM (Support Vector Machine). Two commonly used Bengali printed fonts and three contemporary handwritten databases are used for collecting subjective opinion scores from human readers/writers. On this corpus, which contains the perceptive ground-truth opinion of reading/writing complications, we have undertaken to conduct the experiments. The experimental results obtained on various types of conjunct characters are promising.

In a construction engineering document the sheet number represents a major identifier of the documents. Architecture, Engineering, and Construction (AEC) industry deals with many high dimensional drawing documents. For a construction document management system, it is necessary to extract sheet number from the Title block of the document. Architects and engineers also used to refer another document when creating some new ones. Otherwise, it is very difficult and time consuming to navigate through different files in an interactive way. This paper describes some hand-crafted and Long Short-Term Memory (LSTM) neural-net based features to extract sheet number from the Title Block. At first, we extract the content information from the document. From the extracted content information, we automatically find the sheet number which refers to another document. Then we created the hyperlinks in order to enable the engineers to quickly navigate between files. This work has strong application potential in the AEC industry. We have achieved overall accuracy of 98.52 %.

In AEC (Architectural, Engineering and Constructional) domain, the architects use several types of drawing documents. These documents consist of various category of graphical objects and such objects can be classified based on their shape and curvature properties. Here, we have proposed a query enabled image searching system which can be used to locate objects in a set of documents. To achieve our goal, at-first we have identified a key point search window based on the inflection points of the contour of the object. Then, for each window we try to classify the curve segment into positive or negative class based on the curvature values. Then depending on the predicted class, we get the maxima or minima point on the contour. Now, we use these points as seed values (initial key points) to our feature detection mechanism. To extract the shape and/or curvature properties of the objects at different levels of scale and orientation, we have employed two feature extraction mechanisms. To get the local feature descriptor, we have introduced a neighborhood-based curvature extraction methodology and to consider the global shape of the object, Hu moments based shape descriptor of the object contour is computed. Finally, these two features are normalized, and their combined feature vector is used to compare and measure the similarity between the key points of two objects. We have also introduced a mechanism to identify the query object as a whole, when it consists of multiple overlapping contours for a single object. We have tested our system on a variety of AEC class of drawing documents and the results are quite encouraging. The proposed system will enable the users to locate similar objects inside a project consisting of hundreds of documents. Also, the method can be used for object hyperlinking with some alterations.

The texture classification is a significant problem in the area of pattern recognition. This work proposes a novel Affine Differential Local Mean ZigZag Pattern (ADLMZP) descriptor for texture classification. The proposed method has two manifolds: first Local Mean ZigZag Pattern (LMZP) map is calculated by thresholding the 3 × 3 patch neighbor intensity values with respect to path mean but in a ZigZag weighting fashion, which provides a well discriminated descriptor compared to other local binary descriptors. The local micropattern is obtained by comparing neighbor intensity values with respect to path mean which makes the descriptor robust against noise and illumination variations. Secondly, in order to make it invariant to affine changes, we incorporated an affine differential transformation along with affine gradient magnitude information of a texture image which is differed from Euclidean Gradient. The final ADLMZP descriptor is generated by concatenating the histograms of all Affine Differential Local Mean ZigZag maps. The results are computed over well known KTH-TIPS, Brodatz, and CUReT texture datasets and compared with the state-of-the-art texture classification methods.

Due to the presence of high volume of noise in text documents, it becomes very difficult to achieve high accuracy while performing text line segmentation. Hence, approaches which are dependent on the performance of the line segmentation stage, also suffers. Also due to the variability of noise patterns, the text contents get distorted and sometimes it results in broken strokes of a character. Thus, character level recognition becomes a challenging task for noisy documents. Our proposed method is aimed to address the aforesaid challenges. Since, we are considering the input image to be of noisy kind, a convolutional neural network (CNN) architecture is introduced to initially identify whether the input image contains noise or not. We consider this as a two-class problem and try to identify the respective classes, i.e. noisy or clean. Then, we apply the proposed piecewise projection profile feature and an adaptive region growing based two-stage algorithmic approach which initially identifies the text line upper and lower boundaries in a noisy text document image and then try to regroup the broken strokes of a character to enhance the character recognition accuracy. The proposed method has been tested on a large size dataset containing noisy documents and also different quality measures are computed to establish the effectiveness of the proposed method. From the measures, it can be noted that the text line segmentation accuracy is comparable to other similar state-of-the-art results.

This paper deals with computer-based cognitive analysis towards legibility and aesthetics of a handwritten document. The legible text creates a human perception that the writing can be read effortlessly because of its orthographic clarity. The aesthetic property relates to the beautiful appearance of a handwritten document. In this study, we deal with these properties on offline Bengali handwriting. We formulate both legibility and aesthetic analysis tasks as machine learning problems supervised by the human cognitive system. We employ automatically derived feature-based recurrent neural networks to investigate writing legibility. For aesthetics evaluation, we employ hand-crafted feature-based support vector machines (SVMs). We have collected contemporary Bengali handwritings, on which the subjective legibility and aesthetic scores are provided by human readers. On this corpus containing legibility and aesthetic ground-truth information, we executed our experiments. The experimental results obtained on various handwritings are encouraging.

Architectural and Construction industry deals with many high dimensional paper documents. Architects assign different symbols for referencing other documents for details of a particular section of a large project. The general contractor (GC) or the sub-contractor (SC) will go through these drawings and then shuffle through a pile of documents to navigate to the referenced sheet by reading the information from these symbols. The time taken will increase manifolds if the number of such references is very high. We have proposed a hyperlink based method to automate these manual tasks with a good accuracy and efficiency. This work has strong application potential in the construction industry for aiding in the deadline intensive projects.

The presence of struck-out text in handwritten manuscripts may affect the accuracy of automated writer identification. This paper presents a study on such effects of struck-out text. Here we consider offline English and Bengali handwritten document images. At first, the struck-out texts are detected using a hybrid classifier of a CNN (Convolutional Neural Network) and an SVM (Support Vector Machine). Then the writer identification process is activated on normal and struck-out text separately, to ascertain the impact of struck-out texts. For writer identification, we use two methods: (a) a hand-crafted feature-based SVM classifier, and (b) CNN-extracted auto-derived features with a recurrent neural model. For the experimental analysis, we have generated a database from 100 English and 100 Bengali writers. The performance of our system is very encouraging.

Named entity recognition is an important topic in the field of natural language processing, whereas in document image processing, such recognition is quite challenging without employing any linguistic knowledge. In this paper we propose an approach to detect named entities (NEs) directly from offline handwritten unstructured document images without explicit character/word recognition, and with very little aid from natural language and script rules. At the preprocessing stage, the document image is binarized, and then the text is segmented into words. The slant/skew/baseline corrections of the words are also performed. After preprocessing, the words are sent for NE recognition. We analyze the structural and positional characteristics of NEs and extract some relevant features from the word image. Then the BLSTM neural network is used for NE recognition. Our system also contains a post-processing stage to reduce the true NE rejection rate. The proposed approach produces encouraging results on both historical and modern document images, including those from an Australian archive, which are reported here for the very first time.

In construction or manufacturing industry, engineering drawings are used as blueprint or plan documents to facilitate the construction or manufacturing process. A fairly large construction project involves very large number of these documents, divided into different sub-sections. An engineer or architect often needs to refer different documents while preparing a new one or marking some irregularity in some document. Therefore they need to navigate through different files. It becomes an extremely difficult and time consuming task to move from one file to another in an interactive way. This paper describes an automated technique to access information from the existing drawing documents and create hyperlinks in order to enable the engineers to quickly navigate between files. The overall accuracy of our system for a class of documents is a decent 94.46%.

This paper deals with offline handwritten word recognition of a major Indic script: Bengali. Due to the structure of this script, the characters (mostly ortho-syllables) are frequently overlapping and hard to segment, especially when the writing is cursive. Individual character recognition and the combination of outputs can increase the likelihood of errors. Instead, a better approach can be sending the whole word to a suitable recognizer. Here we use the Convolutional Neural Network (CNN) integrated with a recurrent model for this purpose. Long short-term memory blocks are used as hidden units. Also, the CNN-derived features are employed in a recurrent model with a CTC (Connectionist Temporal Classification) layer to get the output. We have tested our method on three datasets: (a) a publicly available dataset, (b) a new dataset generated by our research group and (c) an unconstrained dataset. The dataset (a) contains 17,091 words, while our dataset (b) contains 107,550 number of words in total. In addition to these, the dataset (c) is comprised of 5,223 words. We have compared our results with those of some earlier work in the area and have found improved performance, which is due to the novel integration of CNNs with the recurrent model.

This paper deals with identifying a writer from his/her offline handwriting. In a multilingual country where a writer can scribe in multiple scripts, writer identification becomes challenging when we have individual handwriting data in one script while we need to verify/identify a writer from handwriting in another script. In this paper such an issue is addressed with two scripts: English and Bengali. Here we model the task as a classification problem, where training data contains only Bengali handwritten samples and testing is performed on English handwritten texts. This work is based on the understanding that a writer has some inherent stroke characteristics that are independent of the script in which (s)he writes. In this work, some implicit structural and statistical features are extracted, and multiple classifiers are employed for writer identification. Many training sessions are run on a database of 100 writers and the performances are analyzed. We have obtained encouraging results on this database, which show the effectiveness of our method.

Writer identification is an essential component in computational forensic. In this paper, we attempt to do this job based only on isolated characters and numerals. For that, at first, some points of interest (keypoints) on the image are detected by structural analysis and SIFT based detector. Then we calculate a set of features within a certain neighborhood of the keypoint and employ fusion rule on multiple probabilistic SVM classifiers output for writer identification. For experimental analysis, a database containing 212,300 isolated Bangla orthosyllabic characters and numerals are generated with the help of 100 writers. We obtain fairly good result to identify a writer. We also try to find a small set of highly discriminative characters storing extra information about the writing style of an individual.

This paper presents a pioneering attempt for developing a recurrent neural net based connectionist system for unconstrained Bengali offline handwriting recognition. The major challenge in configuring such a classification system for a complex script like Bengali is to effectively define the character classes. A novel way of defining character classes is introduced making the recognition problem suitable for using a recurrent model. Indeed, it has to deal with more than nine hundred character classes for which the occurrence probability is very skewed in the language. An off-the-shelf BLSTM-CTC recognizer is used. An open-source dataset is developed for unconstrained Bengali offline handwriting recognition. The dataset contains 2,338 handwritten text lines consisting of about 21,000 word. Experiment shows that with the new definition of character classes the BLSTM-CTC provides an impressive performance for unconstrained Bengali offline handwriting recognition. The character level recognition accuracy is 75.40% without doing any post-processing on the BLSTM-CTC output. Among the 24.60% character level errors, the substitution, deletion and insertion errors are 18.91%, 4.69% and 0.98%, respectively.

Automatic feature extraction plays a pivotal role in defining the overall performance of any Document Image Analysis system, which conventionally operates directly over uncompressed images, although most of the real time systems such as fax machines, digital libraries and e-governance applications accrue and archive the documents in the compressed form for the sake of storage and transfer efficiencies. However, this infers that the compressed documents need to be decompressed before carrying out any operation or analysis which warrants additional computing resources. This limitation in existing systems instigates motivation to explore for feature extraction techniques directly from the compressed documents and eventually design a document analysis system that works directly in compressed domain. Therefore, this research work proposes to extract novel correlation-entropy features directly from run-length compressed TIFF documents. Further, the research work also investigates different methods to demonstrate some of the straight forward application of the proposed features in carrying out compressed document image analysis such as text and non-text component detection, and subsequently performing compressed text line segmentation and characterization, all carried out in the compressed version of the printed text document without going through the stage of decompression. Finally, the experimental results reported validate the developed algorithms and also illustrate that the proposed features are quite powerful in distinguishing compressed text and non-text components.

Segmentation of a text document into lines, words and characters is an important objective in application like OCR and related analytics. However in today's scenario, the documents are compressed for archival and transmission efficiency. Text segmentation in compressed documents warrants decompression, and needs additional computing resources. In this backdrop, the paper proposes a method for text segmentation directly in run-length compressed, printed English text documents. Line segmentation is done using the projection profile technique. Further segmentation into words and characters is accomplished by tracing the white runs along the base region of the text line. During the process, a run based region growing technique is applied in the spatial neighborhood of the white runs to trace the vertical space between the characters. After detecting the character spaces in the entire text line, the decision of word space and character space is made by computing the average character space. Subsequently based on the spatial position of the detected words and characters, their respective compressed segments are extracted. The proposed algorithm is tested with 1083 compressed text lines, and F-measure of 97.93% and 92.86% respectively for word and character segmentation are obtained. Finally an application of word spotting is also presented.

Optical character recognition (OCR) quality, especially for under-resourced scripts like Bangla, as well as for documents printed in old typefaces, is a major concern. An efficient and effective pipeline for OCR betterment is proposed here. The method is unsupervised. It employs a baseline OCR engine as a black box plus a dataset of unlabeled document images. That engine is applied to the images, followed by a visual encoding designed to support efficient word spotting. Given a new document to be analyzed, the black-box recognition engine is first applied. Then, for each result, word spotting is carried out within the dataset. The unreliable OCR outputs of the retrieved word spotting results are then considered. The word that is the centroid of the set of OCR words, measured by edit distance, is deemed a candidate reading.

A handwritten document may contain strike-through texts. If such texts are fed into an OCR system, the output will be garbage. In this paper, we propose a scheme to detect such strike-through texts/words. Using a graph based model, we represent a textual connected component as a graph. The start/end and intersection points of the ink-strokes of a component are marked as graph nodes. There exists an edge between two nodes if they are connected by object (ink) pixels. By eliminating parallel edges and self loops we obtain a simple, undirected, edge-weighted graph of the text-component. The edge-weight is found by adding horizontal/vertical moves weighted by 1 and diagonal moves weighted by √2. In this graph, we find the shortest path which is nearly as long as the width of the text component and maintains a reasonable degree of straightness. This path, if exist, is identified as the strike-through line. Here we deal with handwritten documents in English, Bengali and Devanagari script. Our approach delivers fairly good results.

Automatic recognition of handwritten texts in video lectures has important applications. In video lectures, the presenter usually writes on white / colored board. The video camera often captures the writing board along with certain other objects possibly including the presenter itself. Recognition of handwritten texts from such a video frame requires prior detection of the region of texts in the frame. In this article, we present our recent study of text localization in such video lecture frames. Here, we use Scale Invariant Feature Transform (SIFT) descriptors densely over the entire region of the frame. The descriptors are located on a regular grid of 5 pixels following the usual practice and considered a uniform patch size of 60 × 60 pixels as its support on the basis of an empirical study. This SIFT descriptor at each location (grid point) is fed as a 128-dimensional input feature vector to a Multilayer Perceptron (MLP) network which gives response for each grid point as either text or non-text. Depending on certain aggregate response at each pixel we localize text regions in the input video frame. Next, we employ K-means clustering to detect the text components present in the localized region of the video frame. Finally, two simple rules are applied to decide certain possible detected text components as noise. We obtained encouraging simulation results of this approach on a variety of video lecture frames.

Handwritten character recognition is still a research challenge in OCR discipline, especially for Indian scripts. This paper deals with handwritten Malayalam, a major Indian script, where past work considered a small subset of characters only. In this paper we deal with complete set of basic characters, vowel and consonant signs and compound characters that may be present in the script. Here the recognition task is more difficult because of large numbers of classes and high interclass similarity. To tackle the problem, we proposed a two-stage approach. The first stage is a group classifier, where a group consists of similar characters and those that misclassify among themselves. In the second stage, a character assigned to a group in the first stage is classified to a particular character class. The proposed scheme is more accurate and efficient than a single stage scheme.

Since OCR engines are usually script-dependent, automatic text recognition in multi-script document requires a pre-processor module that identifies the scripts. Based on this motivation, in this paper, we present a word level handwritten Indian script identification technique. To handle this, words are first segmented by morphological dilation and performed connected component labelling. We then employ the Radon transform, discrete wavelet transform, statistical filters and discrete cosine transform to extract the directional multi-resolution spatial features. We tested the features by using linear discriminant analysis, support vector machine and K-nearest neighbour classifiers over 11 different major Indian scripts (including Roman) in bi-script and tri-script scenario. In our tests, we have achieved maximum accuracies of 98% and 96% for bi-script and tri-scipt respectively.

This article deals with binarization of degraded document images. In the proposed approach, Canny edge image of the input degraded document image is obtained after blurring it with a Gaussian filter. Next, the gray values of the two pixels of the input image at the left and right of each edge pixel are noted to form a histogram of these gray values which possesses two distinct peaks and the lowest valley between them provides the global threshold value. Each pixel with gray value greater than the above threshold is turned as background pixel. A small square window is considered around each non-background pixel and certain simple statistics are computed on the gray values of the pixels of this small window based on which the said pixel is turned either background or foreground. Such a local thresholding method at the latter stage can efficiently handle various degradations in the document. The binarized image so obtained is finally subjected to certain common post-processing operations. The proposed method has been compared with a few existing binarization techniques.

Word spotting in graphical documents is a very challenging task. With an increase usage of electronic media, we are in a need of searching objects in graphical documents by some labeled text. To address such scenarios we propose a word spotting system dedicated to graphical documents with Bangla and English scripts. In our proposed system, first text-graphics layers are separated using Gabor filter. In the text layer, character segmentation approach is applied using water reservoir based method to extract each character from the document. Then recognition of these isolated characters is done using rotation invariant feature, coupled with SVM classifier. Well recognized characters are then grouped based on their sizes. Initial spotting is started to find a query word among those groups of characters. In case if the system could spot a word partially due to any noise, SIFT is applied to identify missing portion of that partial spotting. Experimental results on English and Bangla script document images show that the method is feasible to spot a location in text labeled graphical documents.

An embedded solution for automatic detection of Vehicle Identification Numbers (VIN) captured by a mobile camera has a number of real world applications. But the performance of available open source Optical Character Recognition (OCR) systems on VIN images captured by mobile phones is extremely poor because of the image quality affected by various noises. In a recent study of such images, we have observed that the performance of existing open source OCR systems can be improved by applying several image enhancement techniques on these images before sending them to the OCR engine. In this article, we have presented such a method that improves the recognition accuracy from 5.89% up to 82.3%.

In this paper, we describe an approach to distinguish between hand-written text and machine-printed text from annotated machine-printed Bangla Documents images. In applications involving OCR, distinction of machine-printed and hand-written characters is important, so that they can be sent to separate recognition engines. Identification of hand-written parts is useful in deleting those parts and cleaning the document image as well. In this paper a classification system is presented which takes a connected component in the document image and assigns them to two classes namely "machine-printed" and for "hand-written" classes, respectively. The proposed system contains a preprocessing step, which smoothes the object border and finds the Connected Component. Bangla script specific features are extracted from that Connected Component image, and a standard classifier based on SVM generates the final response. Experimental results on a data set show that the proposed approach achieves an overall accuracy of 96.49%.

A major preprocessing step in a multi-script OCR is to identify the script type of the test document image. The published papers on script identification usually assume that the test image is in correct i.e. 0° orientation. But by mistake a document may be fed to the system in wrong orientation, say at an angle of nearly 180° or ±90°. In this method we propose a script identification method that works for unknown orientation for all 11 official Indian scripts. Here, we first find the skew and counter-rotate the document by the skew angle. This will lead to correct (0°) or upside down (180°) orientation. Then script identification is done by a multi-stage tree classifier using features invariant to 0°/180° orientation. Next we go to find the orientation of the image by a two class classifier for each script. Performance of the proposed method has been tested on a variety of documents and promising results have been obtained.

For automatic recognition of Bangla script, only a few studies are reported in the literature, which is in contrast to the role of Bangla as one of the world's major scripts. In this paper we present a new approach to online Bangla handwriting recognition and one of the first to consider cursively written words instead of isolated characters. Our method uses a sub-stroke level feature representation of the script and a writing model based on hidden Markov models. As for the latter an appropriate internal structure is crucial, we investigate different approaches to defining model structures for a highly compositional script like Bangla. In experimental evaluations of a writer independent Bangla word recognition task we show that the use of context-dependent sub-word units achieves quite promising results and significantly outperforms alternatively structured models.

Preprocessing in handwritten text OCR involves line, word and character segmentation. This paper deals with text line identification of handwritten Indian scripts, especially of Bangla, as well as English, Hindi, Malayalam, etc. Here, a new dual method based on interdependency between text-line and inter-line gap is proposed. The method draws curves simultaneously through the text and inter-line gap points found from strip-wise histogram peaks and inter-peak valleys. The curves start from left and move right while one type of points guides the curve of other type so that the curves do not intersect. Then these curves are allowed to iteratively evolve so that the text-line curves cross more character strokes while inter-line curves cross less character strokes and yet keep the curves as straight as possible. After several iterations, the curves stabilize and define the final text-lines and inter-line gaps. The approach works well on text of different scripts with various geometric layouts, including poetry.

In this paper, we present a system towards Indian postal automation based on pin-code and city name recognition. Here, at first, using Run Length Smoothing Approach (RLSA), non-text blocks (postal stamp, postal seal, etc.) are detected and using positional information, Destination Address Block (DAB) is identified from postal documents. Next, lines and words of the DAB are segmented. In India, the address part of a postal document may be written by a combination of two scripts: Latin (English) and a local (State/region) script. It is very difficult to identify the script by which pin-code part is written. To overcome this problem on pin-code part, we have used a two-stage artificial neural network based general scheme to recognize pin-code numbers written in any of the two scripts. To identify the script by which a word/city name is written, we propose a water reservoir concept based feature. For recognition of city names, we propose an NSHP-HMM (Non-Symmetric Half Plane-Hidden Markov Model) based technique. At present, the accuracy of the proposed digit numeral recognition module is 93.14% while that of city name recognition scheme is 86.44%.

This article primarily concerns the problem of isolated handwritten numeral recognition of major Indian scripts. The principal contributions presented here are (a) pioneering development of two databases for handwritten numerals of two most popular Indian scripts, (b) a multistage cascaded recognition scheme using wavelet based multiresolution representations and multilayer perceptron classifiers and (c) application of (b) for the recognition of mixed handwritten numerals of three Indian scripts Devanagari, Bangla and English. The present databases include respectively 22,556 and 23,392 handwritten isolated numeral samples of Devanagari and Bangla collected from real-life situations and these can be made available free of cost to researchers of other academic Institutions. In the proposed scheme, a numeral is subjected to three multilayer perceptron classifiers corresponding to three coarse-to-fine resolution levels in a cascaded manner. If rejection occurred even at the highest resolution, another multilayer perceptron is used as the final attempt to recognize the input numeral by combining the outputs of three classifiers of the previous stages. This scheme has been extended to the situation when the script of a document is not known a priori or the numerals written on a document belong to different scripts. Handwritten numerals in mixed scripts are frequently found in Indian postal mails and table-form documents.

This paper presents an end-to-end administrative document analysis system. This system uses case-based reasoning in order to process documents from known and unknown classes. For each document, the system retrieves the nearest processing experience in order to analyze and interpret the current document. When a complete analysis is done, this document needs to be added to the document database. This requires an incremental learning process in order to take into account every new information, without losing the previous learnt ones. For this purpose, we proposed an improved version of an already existing neural network called Incremental Growing Neural Gas. Applied on documents learning and classification, this neural network reaches a recognition rate of 97.63%.

This paper deals with incremental classification and its particular application to invoice classification. An improved version of an already existant incremental neural network called IGNG (incremental growing neural gas) is used for this purpose. This neural network tries to cover the space of data by adding or deleting neurons as data is fed to the system. The improved version of the IGNG, called I2GNG used local thresholds in order to create or delete neurons. Applied on invoice documents represented with graphs, I2GNG shows a recognition rate of 97.63%.

We describe here a novel scheme for recognition of online handwritten basic characters of Bangla, an Indian script used by more than 200 million people. There are 50 basic characters in Bangla and we have used a database of 24,500 online handwritten isolated character samples written by 70 persons. Samples in this database are composed of one or more strokes and we have collected all the strokes obtained from the training samples of the 50 character classes. These strokes are manually grouped into 54 classes based on the shape similarity of the graphemes that constitute the ideal character shapes. Strokes are recognized by using hidden Markov models (HMM). One HMM is constructed for each stroke class. A second stage of classification is used for recognition of characters using stroke classification results along with 50 look-up-tables (for 50 character classes).

In this paper, we propose to give tools for Bangla handwriting recognition. We present a mechanism to segment documents into text lines and words, and more specifically to detect headline position in each word. Indeed, this headline is an horizontal line on the upper part of most of characters, which is characteristic of Bangla writing. Its localisation is a new approach that can improve text recognition quality. This headline is detected into words inside text lines thanks to a notion of perceptive vision: at a certain distance, text lines appear as line-segments that give the global orientation of words. Watching closer may help to give the exact position of the headline. Consequently, this work is mainly based on applying a segment extractor at different image resolutions and combining extracted information in order to compute the headlines. Our line-segment extractor is based on Kalman filtering.

This paper deals with automatic recognition of offline handwritten Bangla characters. Bangla is the second most popular script among SAARC countries. A new class of features based on Curvelet transform has been used in our classification scheme. The classifier used was SVM with one-against-rest class model. The training and test set were morphologically deformed to get five versions of the same character and each version has been subject to individual SVM classifier. Five classifier outputs obtained in this way have been combined by simple majority voting scheme. The overall recognition accuracy of 95.5% has been obtained on the data set. It is hoped that the Curvelet transform along with such multi-classifier scheme will be useful in other handwritten character data as well.

This paper describes a portable diagnostic telecardiology system, aimed to benefit the rural people of a third world country like India. The designed system consists of two major blocks; the first one is required to be carried to the patient home, named 'Portable Telecardiology Kit'. The second one, named 'Automated Cardiac Signal Processor', a PC based system, to be permanently placed at the nearest rural health care center. The 'Portable Telecardiology Kit' contains a portable ECG machine, a dedicated microcontroller-based full duplex communication system, both interfaced, in a single enclosure. This kit will convert the ECG signal along with a tag page (containing different information regarding patient's health and history of illness) to a digitized serial electromagnetic wave. The signal may be transmitted over a distance of about 7-10 km through a cordless phone under full duplex mode. The PC-based system, after receiving the transmitted signal from patient site, extract the ECG signal after proper filtering, and then pass the signal through an in-built knowledge-base. Finally, a rule-based rough set decision system is generated for the development of an inference engine for disease identification from these time-plane features. The system will generate a report indicating the preliminary level abnormalities, and the precautions that can be adopted at an early stage. The entire report is then sent to the Portable kit at the patient bed-site.

In handwritten character recognition problem, the input images are often affected by distortions and noise. Thus such images at different resolutions include different variations in the input data. In the present work, we considered wavelet transform to obtain multi-resolution representation of each input character image. At each resolution level, we considered three MLPs with different numbers of nodes in their hidden layers and combined the outputs produced by all the MLPs of the whole ensemble by using weighted sum rule, product rule and majority voting. The set of misclassified samples produced by one combination rule is neither a subset nor a superset of a similar set produced by another rule. So, majority voting has been used for the second and final round to produce final outputs after combining the results of the three combinations of the first stage. The proposed approach produced 99.10% correct recognition rate on the test set of Bangia (a major Indian script) numeral database.

Three image databases of handwritten isolated numerals of three different Indian scripts namely Devnagari, Bangla and Oriya are described in this paper. Grayscale images of 22556 Devnagari numerals written by 1049 persons, 12938 Bangla numerals written by 556 persons and 5970 Oriya numerals written by 356 persons form the respective databases. These images were scanned from three different kinds of handwritten documents - postal mails, job application form and another set of forms specially designed by the collectors for the purpose. The only restriction imposed on the writers is to write each numeral within a rectangular box. These databases are free from the limitations that they are neither developed in laboratory environments nor they are non-uniformly distributed over different classes. Also, for comparison purposes, each database has been properly divided into respective training and test sets.

This paper deals with segmentation and recognition of touching characters appearing in scanned mathematical expressions. The technique is based on multifactorial analysis that integrates several factors determining cut-positions in a touching character image. A predictive algorithm is developed for efficient selection of possible cut-positions for segmenting touching characters. Experiment has been carried out using a test-set of reasonable size and results show that a considerable improvement in recognition accuracy can be achieved with a modest increase in computations.

In this paper, we present a system towards Indian postal automation based on the recognition of pin-code and city name of the postal document. In the proposed system, at first, non-text blocks (postal stamp, postal seal etc.) are detected and destination address block (DAB) is identified from the document. Next, lines and words of the DAB are segmented. Since India is a multi-lingual and multi-script country, the address part may be written by combination of two scripts. To identify the script by which a word is written, we propose a water reservoir based technique. It is very difficult to identify the script by which the pin-code portion is written. So, we have used two-stage artificial neural network (NN) based general classifiers for the recognition of pin-code digits written in English/Bangla. For recognition of city names, we propose an NSHP-HMM (non-symmetric half plane-hidden Markov model) based technique.

This paper aims at automatic understanding of online handwritten mathematical expressions (MEs) written on an electronic tablet. The proposed technique involves two major stages: symbol recognition and structural analysis. Combination of two different classifiers have been used to achieve high accuracy for the recognition of symbols. Several online and offline features are used in the structural analysis phase to identify the spatial relationships among symbols. A context-free grammar has been designed to convert the input expressions into their corresponding T(E)X strings which are subsequently converted into MathML format. Contextual information has been used to correct several structure interpretation errors. A new method for evaluating performance of the proposed system has been formulated. Experiments on a dataset of considerable size strongly support the feasibility of the proposed system.

In this paper, a new compression scheme is presented for Indian Language (IL) textual document images. Since OCR technology for IL scripts is not matured enough, transcription of these documents into digital domain needs new techniques that achieve high degree of compression as well as suitable methods to perform various operations like document indexing, retrieval, etc. The proposed method is essentially based on symbolic compression technique, which has been realized with an efficient segmentation-based clustering approach. A soft pattern-matching technique has been implemented using two different feature sets that co-operate each other to build an efficient prototype library. Experiments have been done for documents printed in Devnagari (Hindi) and Bangla scripts, two mostly used script in Indian sub-continent. Test results show that the proposed technique outperforms several standard methods like CCITT Group-4, JBIG, etc. which are frequently used for compression of document images.

This paper proposes a simple voting scheme for off-line recognition of handprinted numerals. One of the main features of the proposed scheme is that this is not script dependent. Another interesting feature is that it is sufficiently fast for real-life applications. In contrast to the usual practices, here we studied the efficiency of a majority voting approach when all the classifiers involved are multilayer perceptrons (MLP) of different sizes and respective features are based on wavelet transforms at different resolution levels. The rationale for this approach is to explore how one can improve the recognition performance without adding much to the requirements for computational time and resources. For simplicity and efficiency, in the present work, we considered only three coarse-to-fine resolution levels of wavelet representation. We primarily simulated the proposed technique on a database of off-line handprinted Bangla (a major Indian script) numerals. We achieved 97.16% correct recognition rate on a test set of 5000 Bangla numerals. In this simulation we used two other disjoint sets (one for training and the other for validation purpose) of sizes 6000 and 1000 respectively. We have also tested our approach on MNIST database for handwritten English digits. The result is comparable with state-of-the-art technologies.

This paper aims at automatic recognition of online handwritten mathematical expressions written on an electronic tablet. The proposed technique involves two major stages: symbol recognition and structural analysis. A multiple-classifier consists of both parametric and nonparametric classifier has been used for recognition of symbols. Parametric classifier is based on hidden Markov model (HMM), whereas, non-parametric classifier uses Nearest Neighbor classification scheme. Structural analysis uses several online and offline features to identify the spatial relationships among symbols. A context free grammar has been designed to convert input expressions into their corresponding Latex strings. Contextual information has been used to correct several errors occurring at both recognition and structural analysis stage. A new method for evaluating performance of the proposed systems has been formulated. Experiments on a dataset of considerable size show high efficiency of the proposed system.

Recognizing mathematical expressions from document image is a key problem in automatic conversion of scientific documents into electronic form. In this paper, we propose a simple grammar-based approach to recognize complex two-dimensional structures of printed mathematical expressions with high accuracy. The proposed technique is based on the structural information of symbols in an expression. An efficient implementation of the grammar is presented. The system generates a TEX string for the input expression. A new criterion for defining structural complexity of a mathematical expression has been formulated to measure the performance of the proposed technique. Experiment using a good representative sample of mathematical expressions shows a reasonably high efficiency of the system.

A document page may contain two or more different scripts. For Optical Character Recognition (OCR) of such a document page, it is necessary to separate different scripts before feeding them to their individual OCR system. In this paper an automatic scheme is presented to identify text lines of different Indian scripts from a document. For the separation task at first the scripts are grouped into a few classes according to script characteristics. Next feature based on water reservoir principle, contour tracing, profile etc. are employed to identify them without any expensive OCR-like algorithms. At present, the system has an overall accuracy of about 97.52%. 1.

How to select a structuring element for a given task is one of the most frequently asked questions in morphology. The present work tries to find a solution for a restricted class of problems, in the domain of shape classification. In this work an algorithm that extracts distinctive structure of each of a given set of objects, which can be used as the structuring elements for object classification system employing the hit-and-miss transformation, is proposed. The proposed algorithm is based on a new measure of local shape property. The method is used to develop a system for Bengali numeral recognition.

This paper proposes a novel approach to automatic recognition of handprinted Bangla (an Indian script) numerals. A modified Topology Adaptive Self-Organizing Neural Network is proposed to extract a vector skeleton from a binary numeral image. Simple heuristics are considered to prune artifacts, if any, in such a skeletal shape. Certain topological and structural features like loops, junctions, positions of terminal nodes, etc. are used along with a hierarchical tree classifier to classify handwritten numerals into smaller subgroups. Multilayer perceptron (MLP) networks are then employed to uniquely classify the numerals belonging to each subgroup. The system is trained using a sample data set of 1800 numerals and we have obtained 93.26% correct recognition rate and 1.71% rejection on a separate test set of another 7760 samples. In addition, a validation set consisting of 1440 samples has been used to determine the termination of the training algorithm of the MLP networks. The proposed scheme is sufficiently robust with respect to considerable object noise.

The paper deals with an optical character recognition system for printed Oriya, a popular Indian script. The development of OCR for this script is difficult because a large number of characters have to be recognized. In the proposed system, the digitized document image is first passed through preprocessing modules like skew correction, line segmentation, zone detection, word and character segmentation, etc. These modules have been developed by combining some conventional techniques with some newly proposed ones. Next, individual characters are recognized using a combination of stroke and run-number based features, along with features obtained from the concept of a water reservoir. The feature detection methods are simple and robust. A prototype of the system has been tested on a variety of printed Oriya material, and currently achieves 96.3% character level accuracy on average.

Existence of touching characters in scanned documents is a major problem in designing an effective character segmentation procedure for OCR systems. In this paper, new techniques are presented for identification and segmentation of touching characters. The techniques are based on fuzzy multifactorial analysis. A predictive algorithm is developed for effectively selecting cut-points to segment touching characters. Initially, our proposed method has been applied for segmenting touching characters that appear in Devnagari (Hindi) and Bangla, two major scripts in the Indian sub-continent. The results obtained from a test-set of considerable size show that a high recognition rate can be achieved with a reasonable amount of computations.

In a general situation, a document page may contain several scriptforms. For optical character recognition (OCR) of such a document page, it is necessary to separate the scripts before feeding them to their individual OCR systems. An automatic technique for the identification of printed Roman, Chinese, Arabic, Devnagari and Bangla text lines from a single document is proposed. Shape based features, statistical features and some features obtained from the concept of a water reservoir are used for script identification. The proposed scheme has an accuracy of about 97.33%.

There are many documents where text lines are not parallel to each other i.e. these lines have different inclinations with the horizontal lines (multi-skew documents). For the OCR of such a document we have to estimate the skew angle of individual text lines because a single rotation cannot de-skew all text lines of the document. In this paper, we describe a robust technique for multi-skew angle detection from Indian documents containing the most popular Indian scripts Devnagari and Bangla. Most characters in these scripts have horizontal lines at the top, called head-lines. The character head-lines usually connect one another in a word and the word appears as a single component. In the proposed method, the connected components are at first labeled and selected. The upper envelopes of selected components are found by column-wise scanning from the top of the component. Portions of the upper envelope satisfying the properties of a digital straight line are detected. They are then clustered into groups belonging to single text lines. Estimates from these individual clusters give the skew angle of each text line. The proposed multi-skew detection technique has an accuracy about 98.3%.

The study of the filamentous objects in waste water has gained momentum due to its significant effect in environmental pollution. The paper describes a neural network based skeleton extraction technique for volumetric images of these biofilm objects. These objects require huge computer storage space. One way to economize the storage space is to represent such images in the form of a vector skeleton (a piecewise linear approximation). Such a skeleton preserves the essential structure of the object. The proposed neural network does not start with a predefined net topology. The topology evolves during the learning process on the basis of the input. The present technique has certain advantages over the conventional 3-D thinning techniques. It achieves data reduction at a higher rate. Also, the proposed technique is highly robust to noise and arbitrary rotations of an image.

We propose an approach for understanding mathematical expressions in printed documents. The overall approach is divided into three main steps: (i) detection of mathematical expressions in a document, (ii) recognition of the symbols present in the expression and (iii) arrangement of the recognized symbols. The detection of mathematical expressions is done through recognition of a few most common symbols and exploiting some structural features of the expressions. A hybrid of feature based and a template-based technique is used for the recognition of symbols. A two-pass approach is used for arrangement of the symbols. The first pass (scanning or lexical analysis) performs a micro-level examination of the symbols in order to identify the symbol groups occurring in them and to determine their categories or descriptors. The second pass (parsing or syntax analysis) processes the descriptors synthesized in the first pass, to determine the syntactic structure of the expression. A set of predefined rules guides the activities in both the passes. Experiments conducted using this approach on a large number of documents show high accuracy.

Segmentation of handwritten words into characters is one of the important components in handwritten text OCR. In this paper we put forward a method for the segmentation of handwritten Bangla (an Indo-Bangladeshi language) text into characters. Based on certain characteristics of Bangla writing methods, different zones across the height of the word are detected. These zones provide certain structural information about the constituent characters of the respective word. In Bangla handwritten texts often there is overlap between rectangular hulls of successive characters. As such the characters are seldom vertically separable. So, we propose a method of recursive contour following in one of the zones across the height of the word to find out the extents within which the main portion of the character lies. If the successive characters are not touching in the zone of contour following, the algorithm gives fairly good results.

Extraction of some meta-information from printed documents without an OCR approach is considered. It can be statistically verified that important terms in articles are printed in italic, bold and all capital style. Detection of these type styles helps in automatic extraction of the lines containing titles, authors' names, subtitles, references as well as sentences having important terms occurring in the text. It also helps in improving the OCR performance for reading the italic text. Some experimental results on the performance of the approach on good quality as well as degraded document images are presented.

In a multi-lingual country like India, a document page may contain more than one script form. Under the three-language formula, the document may be printed in English, Devnagari and one of the other official Indian languages. For OCR of such a document page, it is necessary to separate these three script forms before feeding them to the OCRs of individual scripts. In this paper, an automatic technique of separating the text lines using script characteristics and shape based features is presented. At present, the system has an overall accuracy of about 98.5%.

There are many types of documents where machine-printed and hand-written texts appear intermixed. Since the optical character recognition (OCR) methodologies for machine-printed and hand-written texts are different, it is necessary to separate these two types of text before feeding them to the respective OCR systems. In this paper, we present such a scheme for both Bangla and Devnagari characters. The scheme is based on the structural and statistical features of the machine-printed and hand-written text lines. The classification scheme has an accuracy of about 98.3%.

Describes how the snakes under specified constraints can be used to segment 3-D cells in the volumetric tissue images. A snake contour is initialized in one image slice and the optimum snake contour is used to follow the surface of the cell in different image slices. This method is useful to track the cells which appear to be considerably moved in its lateral position when we see along the depth of the image stack.

This article describes the application of image analysis in evaluation of filamentous bacteria in industrial sludge bulking which may be useful in environmental pollution control. The volumetric images are obtained from confocal microscope. Many conventional and heuristic methods are conveniently used to reduce the noise and enhance the object features. Moreover, it has been shown that the segmentation based on the combination of different image analysis tools is very useful in extracting the filamentous bacterial structure, qualitative and quantitative evaluation of the industrial sludge images and to study the spatial distribution of the filamentous bacteria.

We propose simple and fast algorithms for detection of italic, bold and all-capital words without doing actual character recognition. We present a statistical study which reveals that the detection of such words may play a key role in automatic information retrieval from documents. Moreover, detection of italic words can be used to improve the recognition accuracy of a text recognition system. Considerable number of document images have been tested and our algorithms give accurate results on all the tested images, and the algorithms are very easy to implement.

An efficient LFG parser implementation of Indian languages in general and Bangla in particular has been discussed. It has been shown that the classical technique of nonconfigurational syntactic encoding principles lead to too many disjunctive constraints to be satisfied by the parser. Noting that most of the disjunctions do not exist if an a priori knowledge of the verb is available, a "delayed" syntactic encoding formalism has been proposed. The points of syntactic encoding of noun phrases have been treated a "forward references" that are to be temporarily maintained in a "symbol table" for later precipitation. The proposed solution has two parts. The first part deals with identification of forward reference points, which is done by introducing a new metavariable and augmenting the scope of the Locate operator. The second part deals with precipitation of forward references through special schemata called m-structure schemata projected by the verb. An extension of the solution to one type of Bangla complex sentences has also been proposed (in the Appendix). Implementation notes based on object-oriented programming principles have been provided.

An OCR system is proposed that can read two Indian language scripts: Bangla and Devnagari (Hindi), the most popular ones in the Indian subcontinent. These scripts, having the same origin in ancient Brahmi script, have many features in common and hence a single system can be modeled to recognize them. In the proposed model, document digitization, skew detection, text line segmentation and zone separation, word and character segmentation, character grouping into basic, modifier and compound character category are done for both scripts by the same set of algorithms. The feature sets and classification tree as well as the knowledge base required for error correction (such as lexicon) differ for Bangla and Devnagari. The system shows a good performance for single font scripts printed on clear documents.

In a multi-lingual country like India, a document may contain more than one script forms. For such a document it is necessary to separate different script forms before feeding them to OCRs of individual script. In this paper an automatic word segmentation approach is described which can separate Roman, Bangla and Devnagari scripts present in a single document. The approach has a tree structure where at first Roman script words are separated using the 'headline' feature. The headline is common in Bangla and Devnagari but absent in Roman. Next, Bangla and Devnagari words are separated using some finer characteristics of the character set although recognition of individual character is avoided. At present, the system has an overall accuracy of 96.09%.

Skew angle detection of scanned documents containing most popular Indian scripts (Devnagari and Bangla) is considered. Most characters in these scripts have horizontal lines at the top, called head lines. The character head lines mostly join one another in a word and the word appears as a single component. In the proposed method the components are at first labeled. The upper envelope of a component is found by columnwise scanning from an imaginary line above the component. Portions of upper envelope satisfying the properties of digital straight line are detected. They are clustered as belonging to single text lines. Estimates from individual clusters are combined to get the skew angle. Apart from accuracy and efficiency, an advantage of the method is that character segmentation and zone detection can be readily done from head line information, which is useful in optical character recognition approaches of these scripts.

Clustering techniques such as K-means and Forgy as well as their improved version ISODATA group data around one seed point for each cluster, It is well known that these methods do not work well if the shape of the cluster is elongated or nonconvex. We argue that for a elongated or nonconvex shaped cluster, more than one seed is needed, In this paper a multiseed clustering algorithm is proposed. A density based representative point selection algorithm is used to choose the initial seed points. To assign several seed points to one cluster, a minimal spanning tree guided novel technique is proposed. Also, a border point detection algorithm is proposed for the detection of shape of the cluster. This border in turn signifies whether the cluster is elongated or not, Experimental results show the efficiency of this clustering technique.

We present a semi-automatic method for extracting the 3D boundary of the cells in a compact tissue cross-section photographed by a confocal microscope. The confocal microscope provides pictures at different depths of the cells which can be considered as the image slices of the tissue section. Segmentation of cell boundary from different image slices and combining them to obtain 3D surface automatically is a difficult task. We have developed an approach where given one segmented image slice, the other image slices can be automatically segmented in a layered approach. The idea is to use the information of the previous segmented image slice for segmenting the current image slice.

In this paper we describe a texture segmentation approach without feature computation based on a multilayer perceptron network (MLP). Thus, the users need not bother about the selection and then computation of feature set and hence real-time segmentation may be possible. The basic motivation of the work is the fact that human vision does not consciously compute features for distinguishing different textures in a scene. A single hidden layer MLP network has been found to be most suitable with heuristically chosen input and hidden layer sizes. A method has been used to speedup the learning of the MLP network. The result of segmentation by a trained network usually results in misclassification in the form of speckles. For the removal of such noise an edge-preserving-noise-smoothing technique is proposed. The final segmentation accuracy is well comparable with that of other existing techniques.

This paper deals with an OCR error detection and correction technique for a highly inflectional language script like Bangla (a major Indian language). This is the first report of its kind. Using two separate lexicons of root words and suffixes, candidate root-suffix pairs of each input word are detected, their grammatical agreement are tested and the root/suffix part in which the error has occurred is noted. The correction is made on the corresponding error part of the input string by a fast dictionary access technique. To do so some alternative strings are generated for an erroneous word. Among the alternative strings, those satisfying grammatical agreement in root-suffix and also having smallest Levenstein-Damerau distance are finally chosen as the correct ones. The system has an accuracy of 75.61%.

Extraction of skeletal shape from a 2D dot pattern is discussed. We use a self-organizing neural network model to get a piecewise linear approximation of a skeleton of the pattern. It is found that even without a proper definition of a skeleton, the proposed algorithm is able to produce skeletons that are quite close to what we intuitively feel it should be. In Kohonen's self-organizing model, the set of processors and their neighbourhoods are fixed. We suggest here some modifications of it in which the set of processors and their neighbourhoods change adaptively.

Road networks are important features of satellite imagery. The main contribution of the present road detection method consists of an effective enhancement technique and an efficient segmentation technique that removes non-road pixels step by step from the image where parameters involved: in each step images are determined by the sensor characteristics (like spatial resolution and spectral range) of the satellite. Also, the segmentation process depends not only on the road contrast but also on the road length. Thus, a low contrast but long road segment does not get removed. We have tested the algorithm on a number of images from IRS and SPOT satellites and the results are satisfactory.

A formalism for lexical projection in a Lexical Functional Grammar based syntactic processing environment, where lexical items may consist of more than one word, has been discussed. It is an extension of an earlier formalism that assumed single-worded lexical entities. It has been shown that traditional approaches of handling multi-worded lexical entities in an LFG environment are not quite suitable for Bangla, the language under study, because these approaches assume configurationality whereas Bangla is non-configurational. A “Supra-Lexical” level of analysis has been proposed and a formalism for such analyses introduced. The proposed formalism consists of two phases—an off-line specification phase and an implementation phase. Some tools that are required, along with the syntax for supra-lexical specification has been introduced with examples. Compilation of the specifications has been discussed.

This paper deals with the problem of recognizing and segmenting textures in images. For this purpose the authors employ a technique based on the fractal dimension (FD) and the multi-fractal concept. Six FD features are based on the original image, the above average/high gray level image, the below average/low gray level image, the horizontally smoothed image, the vertically smoothed image, and the multi-fractal dimension of order two. A modified box-counting approach is proposed to estimate the FD, in combination with feature smoothing in order to reduce spurious regions. To segment a scene into the desired number of classes, an unsupervised K-means like clustering approach is used. Mosaics of various natural textures from the Brodatz album as well as microphotographs of thin sections of natural rocks are considered, and the segmentation results to show the efficiency of the technique. Supervised techniques such as minimum-distance and k-nearest neighbor classification are also considered. The results are compared with other techniques.>

The paper presents a method for detecting linear features in a satellite image. For enhancing these features, a local operation based algorithm is proposed which uses a certain multiplicative score rather than the commonly used additive score. The main contribution is in segmenting the linear features in the enhanced image where an adaptive segmentation technique is proposed on the basis of statistical models. A robust parameter estimation is done for this purpose.

In this paper a complete OCR system is described for documents of single Bangla (Bengali) font. The character shapes are recognized by a combination of template and feature matching approach. Images digitized by flatbed scanner are subjected to skew correction, line, word and character segmentation, simple and compound character separation, feature extraction and finally character recognition. A feature based tree classifier is used for simple character recognition. Preprocessing like thinning and skeletonization is not necessary in our scheme and hence the system is quite fast. At present, the system has an accuracy of about 96%. Also, some character occurrence statistics have been computed to model an error detection and correction technique in the near future.

The problems of 3-D digital topology preservation under binary transformations and 3-D object thinning are considered in this correspondence. First, the authors establish the conditions under which transformation of an object voxel to a non-object voxel, or its inverse does not affect the image topology. An efficient algorithm to detect a simple point has been proposed on the basis of those conditions. In this connection, some other interesting properties of 3-D digital geometry are also discussed. Using these properties and the simple point detection algorithm, the authors have proposed an algorithm to generate a surface-skeleton so that the topology of the original image is preserved, the shape of the image is maintained as much as possible, and the results are less affected by noise.>

Deals with the problem of finding the representative points from a data set /spl subespl Rfrsup 2/. Two algorithms are stated. One of the algorithms can find the local best representative or seed points. The extension of these algorithms for three or more dimensions is also discussed. Experimental results on synthetic and real life data are provided which manifest the utility of these algorithms.>

Fractal dimension is an interesting feature proposed to characterize roughness and self-similarity in a picture. This feature has been used in texture segmentation and classification, shape analysis and other problems. An efficient differential box-counting approach to estimate fractal dimension is proposed in this note. By comparison with four other methods, it has been shown that the authors, method is both efficient and accurate. Practical results on artificial and natural textured images are presented.>

A lexical subsystem that contains a morphological level parser is necessary for processing natural languages in general and inflectional languages in particular. Such a subsystem should be able to generate the surface form (i.e. as it appears in a natural sentence) of a word, given the sequence of morphemes constituting the word. Conversely, and more importantly, the subsystem should be able to parse a word into its constituent morphemes. A formalism which enables the lexicon writer to specify the lexicon of an inflectional language is discussed. The specifications are used to build up a lexical description in the form of a lexical database on one hand and a formulation of derivational morphology, called Augmented Finite State Automata (AFSA), on the other. A compact lexical representation has been achieved, where generation of the surface forms of a word, as well as parsing of a word is performed in a computationally attractive manner. The output produced as a result of parsing is suitable for input to the next stage of analysis in a Natural Language Processing (NLP) environment, which, in our case is based on a generalization of the Lexical Functional Grammar (LFG). The application of the formalism on inflectional Indian languages is considered, with Bengali, a modern Indian language, as a case study.

Fractal dimension is a feature used to characterize roughness and self-similarity in a picture. This feature is used in texture segmentation and classification, shape analysis and other problems. An efficient differential box-counting approach to fractal dimension estimation is proposed and compared with four other methods.>

An improved exact coding scheme for a two-tone digital contour is proposed. It is assumed that the contour is one pel thick and perfectly 8-connected. The basic idea of the scheme is to segment the contour at the position where 90 degree bends occur. Each segment is considered as a unit for coding. For coding convenience, the subsegments belonging to the segment, and runs in each subsegment, are identified. The codeword consists of subwords representing the starting direction of the segment, subsegment identification, number of runs in the subsegments, and number of pels in the runs. The scheme has been tested on several contours, and experimental results are presented and compared with those of the best known scheme. It is seen that the proposed scheme is consistently better than the NDSC (neighboring direction segment coding) scheme.>

An image segmentation scheme based on gray-level thresholding is presented. To reduce errors in misclassification, gray-level histograms are sharpened before thresholding using a gray-level transformation function that also leads to an expression for computing the expected threshold. Three new thresholding methods are proposed that reduce the noise, smooth region boundaries, and preserve connectedness among different parts of objects, and are not expensive.>

The binary tree, quadtree, and octree decomposition techniques are widely used in computer graphics and image processing problems. Here, the techniques are reexamined for pattern recognition and shape analysis applications. It has been shown that the quadtree and octree techniques can be used to find the shape hull of a set of points in space while their n-dimensional generalization can be used for divisive hierarchical clustering. Similarly, an n-dimensional binary tree decomposition of feature space can be used for efficient pattern classifier design. Illustrative examples are presented to show the usefulness and efficiency of these hierarchical decomposition techniques.

A differentiation/enhancement edge detector for noisy situations is proposed. The definition of the gradient has been initiated from the algorithm for finding a border in a binary picture. The advantages of this operator have been compared with those of other widely used operators. A measurement of error in extracting edges by thresholding the gradient has also been suggested, and for the detection of acceptable edges, the optimum threshold is chosen corresponding to the minima in the error function.

A linear degradation model for digital image restoration is considered. It is assumed that the noise is additive, uncorrelated and independent of the degradation process and also of original image. A new objective criterion is used for the least square restoration method proposed here. The criterion is to minimize the correlation between image and noise gray levels. The method can be implemented in the Fourier domain and is computationally attractive. The test results on a chromosome picture are demonstrated for different signal-to-noise power ratios.

The redundancy in computation for median filtering, mean filtering, and point and line detection in two-dimensional images is considered. Fast algorithms are described for median evaluation with 3×3 and 5×5 window sizes. The algorithms are compared with existing ones, and the test result on a picture is given. Extension of the algorithms for mean filtering and point and line detection is also discussed.

The problem of statistical pattern recognition with noisy or imprecise feature measurements is considered. An exact analytical expression is found for the probability of misclassification under this condition, for multiclass multivariate systems. The probability of error exceeds that of the ideal case for the special case of two classes, the a priori conditional probability density functions are assumed to be normal, along with the two cases of feature measurement error, namely normal and uniform probability density functions. Monotonicity of the misclassification probability with measurement error variance is shown. Numerical results are presented for both cases over a workable range of parameters. The study is useful in practical pattern recognition problems.