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.

Real-time applications of handwriting analysis have increased drastically in the fields of forensic and information security because of accurate cues. One of such applications is human age estimation based on handwriting for the purpose of immigrant checking. In this paper, we have proposed a new method for age estimation using handwriting analysis using Hu invariant moments and disconnectedness features. To make the proposed method robust to both ruled and un-ruled documents, we propose to explore intersection point detection in Canny edge images of each input document, which results in text components. For each text component pair, we propose Hu invariant moments for extracting disconnectedness features, which in fact measure multi-shape components based on distance, shape and mutual position analysis of components. Furthermore, iterative k-means clustering is proposed for the classification of different age groups. Experimental results on our dataset and some standard datasets, namely, IAM and KHATT, show that the proposed method is effective and outperforms the state-of-the-art methods.

In this paper, we address the issue of scene text detection in the way of direct regression and successfully adapt an effective semantic segmentation model, DeepLab v3+ [1], for this application. In order to handle texts with arbitrary orientations and sizes and improve the recall of small texts, we propose to extract features of multiple scales by inserting multiple Atrous Spatial Pyramid Pooling (ASPP) layers to the DeepLab after the feature maps with different resolutions. Then, we set multiple auxiliary IoU losses at the decoding stage and make auxiliary connections from the intermediate encoding layers to the decoder to assist network training and enhance the discrimination ability of lower encoding layers. Experiments conducted on the benchmark scene text dataset ICDAR2015 demonstrate the superior performance of our proposed network, named as DeepText, over the state-of-the-art approaches.

In longitudinal electronic health records (EHRs), the event records of a patient are distributed over a long period of time and the temporal relations between the events reflect sufficient domain knowledge to benefit prediction tasks such as the rate of inpatient mortality. Medical concept embedding as a feature extraction method that transforms a set of medical concepts with a specific time stamp into a vector, which will be fed into a supervised learning algorithm. The quality of the embedding significantly determines the learning performance over the medical data. In this paper, we propose a medical concept embedding method based on applying a self-attention mechanism to represent each medical concept. We propose a novel attention mechanism which captures the contextual information and temporal relationships between medical concepts. A light-weight neural net, "Temporal Self-Attention Network (TeSAN)", is then proposed to learn medical concept embedding based solely on the proposed attention mechanism. To test the effectiveness of our proposed methods, we have conducted clustering and prediction tasks on two public EHRs datasets comparing TeSAN against five state-of-the-art embedding methods. The experimental results demonstrate that the proposed TeSAN model is superior to all the compared methods. To the best of our knowledge, this work is the first to exploit temporal self-attentive relations between medical events.

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.

As digitised documents normally contain a large variety of structures, a page segmentation- and layout-free method for document image retrieval is preferable. In this research work, therefore, wavelet transform as a transform-based approach is initially used to provide different under-sampled images from the original image. Then, Gist operator, as a feature extraction technique, is employed to extract a set of global features from the original image as well as the sub-images obtained from the wavelet transform. Moreover, the column-wise variances of the values in each sub-image are computed and they are then concatenated to obtain another set of features. Considering each feature set, locality-sensitive hashing is employed to compute similarity distances between a query and the document images in the database. Finally, a classifier fusion technique using the mean function is taken into account to provide a document image retrieval result. The combination of these features and a clustering score fusion strategy provides higher document image retrieval accuracy. Two different databases of the document image are considered for experimentation. The results obtained from the experimental study are detailed and the results are encouraging.

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.

Printer identification is challenging due to advanced software technologies in the field of forgery detection. This paper presents a new idea of using the Fourier transform residual for the identification of documents printed by different printers. The proposed approach first convolves a Laplacian mask with a Fourier transform in the frequency domain to smoothen the edges. Next, we apply an inverse Fourier transform to reconstruct images from smoothed information (RFL). Similarly, the proposed approach reconstructs images using gray information of the input image (RFG). Then the residual is calculated by subtracting RFG from RFL. The set of statistical features, texture and spatial features are extracted from residual images for printer identification. Experimental results with the existing method on our dataset and a standard dataset show that the proposed approach outperforms the existing approach on both the datasets in terms of classification rate, recall, precision and F-measure.

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 this paper, a full reference document image quality assessment (FR DIQA) method using texture features is proposed. Local binary patterns (LBP) as texture features are extracted at the local and global levels for each image. For each extracted LBP feature set, a similarity measure called the LBP similarity index (LBPSI) is computed. A weighting strategy is further proposed to improve the LBPSI obtained based on local LBP features. The LBPSIs computed for both local and global features are then combined to get the final LBPSI, which also provides the best performance for DIQA. To evaluate the proposed method, two different datasets were used. The first dataset is composed of document images, whereas the second one includes natural scene images. The mean human opinion scores (MHOS) were considered as ground truth for performance evaluation. The results obtained from the proposed LBPSI method indicate a significant improvement in automatically/accurately predicting image quality, especially on the document image-based dataset.

This paper presents a survey on different approaches for removing the marginal noise from document images, and anlaysing the research challenges of those methods relating to handwritten historical datasets. In this survey, historical documents collected from Australian Archives and Libraries are introduced and the associated layout complexities of those document images are also described. Benchmarking other historical databases related to this work is also discussed. This survey discusses the difficulties and suitability of the state-of-the-art methods to remove marginal noise as well as preserving the text content from handwritten historical documents. This survey helps researchers to identify appropriate methods according to the associated marginal noise and also illustrates their drawbacks in order to make suggestions for developing approaches, which are more general and robust for any datasets.

We propose a holistic, dynamic method to preserve text content with zero tolerance while removing marginal noise for historical handwritten document images. The key idea is to identify and analyze the region between the sharp peak at the edge and page frame of the text content at each margin. Depending on the proximity of the sharp peak to the text, the text content is then extracted from the document image. This method automatically adapts thresholds for each single document image and is directly applicable to gray-scale images. The proposed method is evaluated on four diverse handwritten historical datasets: Queensland State Archive (QSA), Saint Gall, Parzival and the Prosecution Project. Experimental results show that the proposed method achieves higher accuracy compared with other methods tested on the Saint Gall and Parzival datasets, whilst for the other two Australian datasets, which have been introduced here for the first time, the results are very encouraging.

In this paper, a signature verification method based on texture features involving off-line signatures written in two different Indian scripts is proposed. Both Local Binary Patterns (LBP) and Uniform Local Binary Patterns (ULBP), as powerful texture feature extraction techniques, are used for characterizing off-line signatures. The Nearest Neighbour (NN) technique is considered as the similarity metric for signature verification in the proposed method. To evaluate the proposed verification approach, a large Bangla and Hindi off-line signature dataset (BHSig260) comprising 6240 (260×24) genuine signatures and 7800 (260×30) skilled forgeries was introduced and further used for experimentation. We further used the GPDS-100 signature dataset for a comparison. The experiments were conducted, and the verification accuracies were separately computed for the LBP and ULBP texture features. There were no remarkable changes in the results obtained applying the LBP and ULBP features for verification when the BHSig260 and GPDS-100 signature datasets were used for experimentation.

This paper presents the results of the ICDAR 2015 competition on signature verification and writer identification for on- and off-line skilled forgeries jointly organized by PR-researchers and Forensic Handwriting Examiners (FHEs). The aim is to bridge the gap between recent technological developments and forensic casework. Two modalities (signatures and handwritten text) are considered and training and evaluation data are collected and provided by FHEs and PR-researchers. Four tasks are defined for four different languages; Bengali off-line signature verification, Italian off-line signature verification, German on-line signature verification, and English handwritten text based writer identification. In total, 40 systems have participated in this competition. The participants of the signatures modality were motivated to report their results in Likelihood Ratios (LRs). This has made the systems even more interesting for application in forensic casework. For evaluating the performance of the systems, we have used the forensically substantial Cost of Log Likelihood Ratios (Ĉllr) in the case of signatures, and the F-measure in the case of handwritten text.

Automatic document interpretation and retrieval is an important task to access handwritten digitized document repositories. In documents, the date is an important field and it has various applications such as date-wise document indexing/retrieval. In this paper a framework has been proposed for automatic date field extraction from handwritten documents. In order to design the system, sliding window-wise Local Gradient Histogram (LGH)-based features and a character-level Hidden Markov Model (HMM)-based approach have been applied for segmentation and recognition. Individual date components such as month-word (month written in word form i.e. January, Jan, etc.), numeral, punctuation and contraction categories are segmented and labelled from a text line. Next, a Histogram of Gradient (HoG)-based features and a Support Vector Machine (SVM)- based classifier have been used to improve the results obtained from the HMM-based recognition system. Subsequently, both numeric and semi-numeric regular expressions of date patterns have been considered for undertaking date pattern extraction in labelled components. The experiments are performed on an English document dataset and the encouraging results obtained from the approach indicate the effectiveness of the proposed system.

This paper presents the final results of the ICDAR 2015 Competition on Video Script Identification. A description and performance of the participating systems in the competition are reported. The general objective of the competition is to evaluate and benchmark the available methods on word-wise video script identification. It also provides a platform for researchers around the globe to particularly address the video script identification problem and video text recognition in general. The competition was organised around four different tasks involving various combinations of scripts comprising tri-script and multi-script scenarios. The dataset used in the competition comprised ten different scripts. In total, six systems were received from five participants over the tasks offered. This report details the competition dataset specifications, evaluation criteria, summary of the participating systems and their performance across different tasks. The systems submitted by Google Inc. were the winner of the competition for all the tasks, whereas the systems received from Huazhong University of Science and Technology (HUST) and Computer Vision Center (CVC) were very close competitors.

Text recognition from video frames is a challenging task due to low resolution, blur, complex and coloured backgrounds, noise, to mention a few. Consequently, the traditional ways of text recognition from scanned documents having simple backgrounds fails when applied to video text. Although there are various techniques available for text recognition from handwritten and printed documents with simple backgrounds, text recognition from video frames has not been comprehensively investigated, especially for multi-lingual videos. In this paper, we present a technique for multi-lingual video text recognition which involves script identification in the first stage, followed by word and character recognition, and finally the results are refined using a post-processing technique. Considering the inherent problems in videos, a Spatial Pyramid Matching (SPM) based technique, using patch-based SIFT descriptors and SVM classifier, is employed for script identification. In the next stage, a Hidden Markov Model (HMM) based approach is used for word and character recognition, which utilizes the context information. Finally, a lexicon-based post-processing technique is applied to verify and refine the word recognition results. The proposed method was tested on a dataset comprising of 4800 words from three different scripts, namely, Roman (English), Hindi and Bengali. The script identification results obtained are encouraging. The word and character recognition results are also encouraging considering the complexity and problems associated with video text processing.

There are only a few studies undertaken in developing automatic assessment systems using handwriting recognition, even though a successful system would undoubtedly benefit the education system as schools and universities in many countries still employ paper-based examinations. To the best of the authors' knowledge, there is no existing work on an automatic off-line short answer assessment system comprising a student identification component. Hence in this paper, the authors propose a system towards this, where a new feature extraction technique called the Enhanced Water Reservoir, Loop and Gaussian Grid Feature, as well as other enhanced feature extraction techniques were utilised. Artificial Neural Networks and Support Vector Machines were employed as the classifiers; they were used for the investigation, and a comparison of the recognition and accuracy rates of the proposed systems, as well as the feature extraction techniques, was undertaken. The proposed assessment system achieved a recognition rate of 87.12% with 91.12% assessment accuracy, and the student identification component obtained a recognition rate of 99.52% with a 100% identification accuracy rate.

This paper presents the results of the ICDAR2013 competitions on signature verification and writer identification for on- and offline skilled forgeries jointly organized by PR researchers and Forensic Handwriting Examiners (FHEs). The aim is to bridge the gap between recent technological developments and forensic casework. Two modalities (signatures, and handwritten text) are considered where training and evaluation data (in Dutch and Japanese) were collected and provided by FHEs and PR-researchers. Four tasks were defined where the systems had to perform Dutch offline signature verification, Japanese offline signature verification, Japanese online signature verification, and Dutch writer identification. The participants of the signatures modality were motivated to report their results in Likelihood Ratios (LR). This has made the systems even more interesting for application in forensic casework. For evaluation of signatures modality, we used both the traditional Equal Error Rate (EER) and forensically substantial Cost of Log Likelihood Ratios (Ĉllr). The system having the smallest value of the Minimum Cost of Log Likelihood Ratio (Ĉllrmin) is declared winner. For evaluation of the handwritten text modality, we used the precision and accuracy measures and winners are announced on the basis of best F-measure value.

Script identification is an essential step for the efficient use of the appropriate OCR in multilingual document images. There are various techniques available for script identification from printed and handwritten document images, but script identification from video frames has not been explored much. This paper presents a study of some pre-processing techniques and features for word-wise script identification from video frames. Traditional features, namely Zernike moments, Gabor and gradient, have performed well for handwritten and printed documents having simple backgrounds and adequate resolution for OCR. Video frames are mostly coloured and suffer from low resolution, blur, background noise, to mention a few. In this paper, an attempt has been made to explore whether the traditional script identification techniques can be useful in video frames. Three feature extraction techniques, namely Zernike moments, Gabor and gradient features, and SVM classifiers were considered for analyzing three popular scripts, namely English, Bengali and Hindi. Some pre-processing techniques such as super resolution and skeletonization of the original word images were used in order to overcome the inherent problems with video. Experiments show that the super resolution technique with gradient features has performed well, and an accuracy of 87.5% was achieved when testing on 896 words from three different scripts. The study also reveals that the use of proper pre-processing approaches can be helpful in applying traditional script identification techniques to video frames.

This paper presents a two-stage method for multi-oriented video character segmentation. Words segmented from video text lines are considered for character segmentation in the present work. Words can contain isolated or non-touching characters, as well as touching characters. Therefore, the character segmentation problem can be viewed as a two stage problem. In the first stage, text cluster is identified and isolated (non-touching) characters are segmented. The orientation of each word is computed and the segmentation paths are found in the direction perpendicular to the orientation. Candidate segmentation points computed using the top distance profile are used to find the segmentation path between the characters considering the background cluster. In the second stage, the segmentation results are verified and a check is performed to ascertain whether the word component contains touching characters or not. The average width of the components is used to find the touching character components. For segmentation of the touching characters, segmentation points are then found using average stroke width information, along with the top and bottom distance profiles. The proposed method was tested on a large dataset and was evaluated in terms of precision, recall and f-measure. A comparative study with existing methods reveals the superiority of the proposed method.

With increasing computational power, researchers in the area of off-line signature verification have been able to investigate feature extraction techniques that produce large-dimensional feature vectors. However, a large feature vector is not necessarily associated with high performance. This paper investigates the performance of a small feature set consisting of 33 feature values. In the experiments using Support Vector Machines (SVMs), an average error rate (AER) of 16.80% was obtained together with a low false acceptance rate (FAR) for random forgeries of 0.19%. The significant reduction of the error rate was obtained when the proposed global features were employed, which demonstrates their astonishingly high discriminant power. These results suggest that the use of global features for the off-line signature verification problem is worth further investigation.

In the field of information security, biometric systems play an important role. Within biometrics, automatic signature identification and verification has been a strong research area because of the social and legal acceptance and extensive use of the written signature as an individual authentication. Signature verification is a process in which the questioned signature is examined in detail in order to determine whether it belongs to the claimed person or not. Despite substantial research in the field of signature verification involving Western signatures, very few works have been dedicated to non-Western signatures such as Chinese, Japanese, Arabic, or Persian etc. In this paper, the performance of an off-line signature verification system involving Bangla signatures, whose style is distinct from Western scripts, was investigated. The Gaussian Grid feature extraction technique was employed for feature extraction and Support Vector Machines (SVMs) were considered for classification. The Bangla signature database employed in the experiments consisted of 3000 forgeries and 2400 genuine signatures. An encouraging accuracy of 90.4% was obtained from the experiments.

Extraction and recognition of text present in video has become a very popular research area in the last decade. Generally, text present in video frames is of different size, orientation, style, etc. with complex backgrounds, noise, low resolution and contrast. These factors make the automatic text extraction and recognition in video frames a challenging task. A large number of techniques have been proposed by various researchers in the recent past to address the problem. This paper presents a review of various state-of-the-art techniques proposed towards different stages (e.g. detection, localization, extraction, etc.) of text information processing in video frames. Looking at the growing popularity and the recent developments in the processing of text in video frames, this review imparts details of current trends and potential directions for further research activities to assist researchers.

Text detection in video frames plays a vital role in enhancing the performance of information extraction systems because the text in video frames helps in indexing and retrieving video efficiently and accurately. This paper presents a new method for arbitrarily-oriented text detection in video, based on dominant text pixel selection, text representatives and region growing. The method uses gradient pixel direction and magnitude corresponding to Sobel edge pixels of the input frame to obtain dominant text pixels. Edge components in the Sobel edge map corresponding to dominant text pixels are then extracted and we call them text representatives. We eliminate broken segments of each text representatives to get candidate text representatives. Then the perimeter of candidate text representatives grows along the text direction in the Sobel edge map to group the neighboring text components which we call word patches. The word patches are used for finding the direction of text lines and then the word patches are expanded in the same direction in the Sobel edge map to group the neighboring word patches and to restore missing text information. This results in extraction of arbitrarily-oriented text from the video frame. To evaluate the method, we considered arbitrarily-oriented data, non-horizontal data, horizontal data, Hua's data and ICDAR-2003 competition data (Camera images). The experimental results show that the proposed method outperforms the existing method in terms of recall and f-measure.

The Netherlands Forensic Institute and the Institute for Forensic Science in Shanghai are in search of a signature verification system that can be implemented in forensic casework and research to objectify results. We want to bridge the gap between recent technological developments and forensic casework. In collaboration with the German Research Center for Artificial Intelligence we have organized a signature verification competition on datasets with two scripts (Dutch and Chinese) in which we asked to compare questioned signatures against a set of reference signatures. We have received 12 systems from 5 institutes and performed experiments on online and offline Dutch and Chinese signatures. For evaluation, we applied methods used by Forensic Handwriting Examiners (FHEs) to assess the value of the evidence, i.e., we took the likelihood ratios more into account than in previous competitions. The data set was quite challenging and the results are very interesting.

Similar to many other pattern recognition problems, feature extraction contributes significantly to the overall performance of an off-line signature verification system. To be successful, a feature extraction technique must be tolerant to different types of variation whilst preserving essential information of input patterns. In this paper, we describe a grid-based feature extraction technique that utilises directional information extracted from the signature contour, i.e. the chain code histogram. Our experimental results for signature verification indicated that, by applying a suitable 2D Gaussian filter on the matrices containing the chain code histograms, an average error rate (AER) of 13.90% can be obtained whilst maintaining the false acceptance rate (FAR) for random forgeries as low as 0.02%. These figures are comparable or better than those reported by other state of the art feature extraction techniques such as the Modified Direction Feature (MDF) and the Gradient feature.

On-line handwriting recognition systems are usually better than their off-line counterparts thanks to the accessibility of dynamic information such as stroke order, velocity, acceleration, and pressure. Whilst the exact value of velocity as well as acceleration or pressure is unlikely to be recoverable, the temporal order of the strokes or the pen trajectory is shown to be more promising for recovery. The published experimental results suggest that the recovered pen trajectory information actually improves the off-line recognition accuracy. This paper presents an overview and discussion of pen trajectory recovery methods developed to date.

Global features based on the boundary of a signature and its projections are described for enhancing the process of automated signature verification. The first global feature is derived from the total psilaenergypsila a writer uses to create their signature. The second feature employs information from the vertical and horizontal projections of a signature, focusing on the proportion of the distance between key strokes in the image, and the height/width of the signature. The combination of these features with the Modified Direction Feature (MDF) and the ratio feature showed promising results for the off-line signature verification problem. When being trained using 12 genuine specimens and 400 random forgeries taken from a publicly available database, the Support Vector Machine (SVM) classifier obtained an average error rate (AER) of 17.25%. The false acceptance rate (FAR) for random forgeries was also kept as low as 0.08%.

As a biometric, signatures have been widely used to identify people. In the context of static image processing, the lack of dynamic information such as velocity, pressure and the direction and sequence of strokes has made the realization of accurate off-line signature verification systems more challenging as compared to their on-line counterparts. In this paper, we propose an effective method to perform off-line signature verification based on intelligent techniques. Structural features are extracted from the signature's contour using the modified direction feature (MDF) and its extended version: the Enhanced MDF (EMDF). Two neural network-based techniques and Support Vector Machines (SVMs) were investigated and compared for the process of signature verification. The classifiers were trained using genuine specimens and other randomly selected signatures taken from a publicly available database of 3840 genuine signatures from 160 volunteers and 4800 targeted forged signatures. A distinguishing error rate (DER) of 17.78% was obtained with the SVM whilst keeping the false acceptance rate for random forgeries (FARR) below 0.16%.

This paper presents an enhanced heuristic segmenter (EHS) and an improved neural-based segmentation technique for segmenting cursive words and validating prospective segmentation points respectively. The EHS employs two new features, ligature detection and a neural assistant, to locate prospective segmentation points. The improved neural-based segmentation technique can then be used to examine the prospective segmentation points by fusion of confidence values obtained from left and centre character recognition outputs in addition to the segmentation point validation (SPV) output. The improved neural-based segmentation technique uses a recently proposed feature extraction technique (modified direction feature) for representing the segmentation points and characters to enhance the overall segmentation process. The EHS and the neural-based segmentation technique have been implemented and tested on a benchmark database providing encouraging results.

High accuracy character recognition techniques can provide useful information for segmentation-based handwritten word recognition systems. This research describes neural network-based techniques for segmented character recognition that may be applied to the segmentation and recognition components of an off-line handwritten word recognition system. Two neural architectures along with two different feature extraction techniques were investigated. A novel technique for character feature extraction is discussed and compared with others in the literature. Recognition results above 80% are reported using characters automatically segmented from the CEDAR benchmark database as well as standard CEDAR alphanumerics.

Analyses the performance of our improved segmentation algorithm tested on the CEDAR benchmark database of handwritten words. Segmentation is achieved through the extraction of a wide range of information adjacent to or surrounding suspicious segmentation points. Initially, a heuristic technique is employed to search for structural features and to over-segment each word. For each segmentation point that is located, the left character (preceding the segmentation point) and centre character (centred on the segmentation point) are extracted along with other features from the segmentation area. The aforementioned features are presented to trained character and segmentation point validation neural networks to evaluate a number of confidence values. Finally, the confidence values are fused to obtain the final segmentation decision. Based on a detailed analysis, it was observed that the left and centre character networks increased the accuracy of the segmentation algorithm.

A new intelligent segmentation technique is proposed that may be used in conjunction with a neural classifier and a simple lexicon for the recognition of difficult handwritten words. A heuristic segmentation algorithm is initially used to over-segment each word. An artificial neural network (ANN) trained with 32,034 segmentation points is then used to verify the validity of the segmentation points found. Following segmentation, character matrices from each word are extracted, normalised and then passed through a global feature extractor, after which a second ANN trained with segmented characters is used for classification. These recognised characters are grouped into words and presented to a variable-length lexicon that utilises a string processing algorithm to compare and retrieve those words with the highest confidences. This research provides promising results for segmentation, character and word recognition.