Human brain is adept at continually acquiring new skills through structured patterns without forgetting previous learning, a feat that neural networks struggle to emulate. When these networks are exposed to new information, they tend to experience a significant decline in performance on tasks they were previously trained on, a phenomenon known as catastrophic forgetting. We introduce a novel approach to representation learning that utilizes a hierarchical structure. In our proposed method, individual classes are encapsulated and dynamically routed to maintain relationships between similar classes at different levels. Our model develops into a tree-like architecture, with each phase adding nodes that contain capsules of semantically related classes. During training, only the nodes representing new classes and their children are trained, while the rest of the tree remains frozen. This effectively preserves the representations of older classes.

Computer vision applications for document image understanding (DIU) such as optical character recognition, word spotting, enhancement etc. suffer from structural deformations like strike-outs and unconstrained strokes, to name a few. They also suffer from texture degradation due to blurring, aging, or blotting-spots etc. The DIU applications with deep networks are limited to constrained environment and lack diverse data with text-level and pixel-level annotation simultaneously. In this work, we propose a generative framework to produce realistic synthetic handwritten document images with simultaneous annotation of text and corresponding pixel-level spatial foreground information. The proposed approach generates realistic backgrounds with artificial handwritten texts which supplements data-augmentation in multiple unconstrained DIU systems. The proposed framework is an early work to facilitate DIU system-evaluation in both image quality and recognition performance at a go.

Hyper-spectral images have hundreds of bands, which provide rich information compared to popular RGB images. This information can be leveraged to accurately segment images into multiple fine-grained classes. However, processing high-dimensional data poses challenges. In this work, we propose two methods for land cover classification using hyper-spectral images, aiming to achieve accurate classification into multiple fine-grained classes. One of the main challenges in land cover classification is the availability of labeled data. To address this, we employ semi-supervised methods that require less labeled data and effectively utilize the vast amount of un-labeled data. We evaluate our methods on the Indian Pine dataset, while varying the amount of labeled pixels used for training.

With the availability of commercial Light Field (LF) cameras, LF imaging has emerged as an up-and-coming technology in computational photography. However, the spatial resolution is significantly constrained in commercial micro-lens-based LF cameras because of the inherent multiplexing of spatial and angular information. Therefore, it becomes the main bottleneck for other applications of light field cameras. This paper proposes an adaptation module in a pre-trained Single Image Super-Resolution (SISR) network to leverage the powerful SISR model instead of using highly engineered light field imaging domain-specific Super Resolution models. The adaption module consists of a Sub-aperture Shift block and a fusion block. It is an adaptation in the SISR network to further exploit the spatial and angular information in LF images to improve the super-resolution performance. Experimental validation shows that the proposed method outperforms existing light field super-resolution algorithms. It also achieves PSNR gains of more than 1 dB across all the datasets as compared to the same pre-trained SISR models for scale factor 2, and PSNR gains 0.6 − 1 dB for scale factor 4.

Existing dynamic saliency prediction models face challenges like inefficient spatio-temporal feature integration, ineffective multi-scale feature extraction, and lacking domain adaptation because of huge pre-trained backbone networks. In this paper, we propose a two pathway architecture with effective feature integration of spatial and temporal domains at multiple scales for video saliency prediction. Frame and optical flow pathways extract features from video frame and optical flow maps, respectively using a series of cross-concatenated multi-scale residual (CMR) blocks. We name this network as two-pathway CMRNet (TP-CMRNet). Every CMR block follows a feature fusion and attention module for merging features from two pathways and guiding the network to weigh salient regions, respectively. A bi-directional LSTM module is used for learning the task by looking at previous and next video frames. We build a simple decoder for feature reconstruction into the final attention map. TP-CMRNet is comprehensively evaluated using three benchmark datasets: DHF1K, Hollywood-2, and UCF sports. We observe that our model performs at par with other deep dynamic models. In particular, we outperform all the other models with a lesser number of model parameters and lower inference time.

Standard pre-trained convolutional neural networks are deployed on different task-specific limited class applications. These applications require classifying images of a much smaller subset of classes than that of the original large domain dataset on which the network is pre-trained. Therefore, a computationally inefficient and over-represented network is obtained. Hierarchically Self Decomposing CNN (HSD-CNN) addresses this issue by dissecting the network into sub-networks in an automated hierarchical fashion such that each sub-network is useful for classifying images of closely related classes. However, visual similarities are not always well-aligned with the semantic understanding of humans. In this paper, we propose a method that aids the pre-trained network to learn the hierarchy of classes derived from standard taxonomy, WordNet and, produce sub-networks corresponding to semantically meaningful classes upon decomposition. Experimental results show that the cluster of classes obtained for each sub-network is semantically closer according to WordNet hierarchy without degradation in overall accuracy.

Though surface mining causes huge land use and land cover changes, it is a widely used technique. Many ores like Iron, Coal, Copper, Dolomite, Diamond, Gold, etc. are extracted using surface mining. In literature, semi supervised and supervised method are used for classification and monitoring of surface mining regions. Spectral indexes to detect these surface mine regions are yet to be defined. An index namely, coal mine index (CMI), has been proposed to detect surface coal mine regions. The paper focuses on performance of CMI in seven different surface mining regions. In this regard, a brief survey on performance and applicability of CMI over these surface mining regions are discussed in this paper.

Existing deep architectures for visual saliency prediction face problems like inefficient feature encoding, larger inference times, and a huge number of model parameters. One possible solution is to make the local and global contextual feature extraction computationally less intensive by a novel lighter architecture. In this work, we propose an end-to-end learnable, inter-scale information sharing residual block based architecture for saliency prediction. A series of these blocks are used for efficient multi-scale feature extraction followed by a dilated inception module (DIM) and a novel decoder. We name this network as cross-concatenated multi-scale residual (CMR) block based network, CMRNet. We comprehensively evaluate our architecture on three datasets: SALICON, MIT1003, and MIT300. Experimental results show that our model works at par with other state-of-the-art models. Especially, our model outperforms all the other models with a smaller inference time and a lesser number of model parameters.

Recently, convolution neural network (CNN)-based hyperspectral image (HSI) classification has gained attention due to its remarkable performance when the number of training samples is sufficiently large. Existing approaches involve 1-D, 2-D or 3D CNN based architecture. 3D CNN involves heavy computation and others are inefficient in using spatial and spectral information jointly. In this work, we propose to use pointwise 3-D convolution to extract spectral feature, followed by 2-D convolution to extract spectral-spatial features jointly in an end to end manner. We have shown a variation in inception-like high level architecture for feature extraction. This is guided by the fact that averaging a large number of feature may loose some unique information. On the other hand, existence of spatial variability within a class and similarity among different classes incurs degradation in HSI classification. To combat with that, we propose a novel target-patch-orientation (TPO) scheme to form a spatial-spectral neighborhood of a pixel. The Experimental results reveal that TPO scheme has positive impact on the proposed model.

Enormous volumes of data made available by the high-resolution satellite imagery enable us to use a deep framework in the field of remote sensing for image classification. Recently, deep learning has been an area of interest for the researchers in the computer vision domain due to its high efficiency toward large-scale, high-dimensional data. In this letter, we propose an unsupervised learning algorithm to cluster hybrid polarimetric SAR images, and dual-polarized SAR images using the deep framework. We use feature extraction layers of the VGG16 model with batch normalization, which is trained with small patches derived from the hybrid polarimetric SAR images. It uses an entropy-based loss function and an adaptive learning rate optimization algorithm, Adam, for training. Broadly, the patches are segmented into three classes, namely, surface, volume, and double-bounce, which are defined with reference to the SAR scattering characteristics. Furthermore, we classify volume into dense forest region and agricultural crop fields. We also observe mixed classes between volume and double-bounce, mainly covering the settlements surrounded by areas covered by tall trees. Furthermore, we use transfer learning for generating the labels for dual-polarized images by using the learned weights of a hybrid polarized image model. Such a technique renders an average accuracy of 89.70% and 86.08% for hybrid polarized SAR images and dual-polarized SAR images, respectively. Hence, this method explores the spatial characteristics of remotely sensed images to distinguish urban settlements, water bodies, agricultural, and forest areas from the underlying scene in an unsupervised fashion.

In recent years, coronary artery disease is escalating and is likely to assume an epidemic proportion by 2030. Currently the reliable methods for detection of coronary arterial lesions are either conventional coronary angiogram (CAG) or MDCT (Multiple Detector Computed Tomography) coronary angiogram. Conventional CAG is an invasive procedure. Conventional CAG and CT (Computed Tomography) angiogram, both require expert supervision of either an interventional cardiologist or a radiologist. In this work, we have proposed a novel design and method for non-invasive detection and localization of coronary arterial lesion using Impedance Cardiography (ICG). The ICG signal recorded by the proposed device is used to extract feature points and compute augmentation index, amplitude and other time related parameters. The extracted features are used as input to a trained artificial neural network, for detection and prediction of coronary arterial lesions. The trained network generates specialized models, to be used for diagnosis of arterial lesions. The proposed methodology detects lesion in Left main coronary artery (LMCA), Left anterior descending artery (LAD), Diagonal branch, Left circumflex artery (LCX), and Right coronary artery (RCA) with an accuracy of 92%, 82%, 76%, 76%, 84% respectively. The proposed device could be also used by a common individual for detection of arterial lesion without any expert supervision, unassisted. The proposed algorithm eliminates the need of CAG for diagnosis of coronary arterial lesions (stenosis), and provides an insight into a new method for non-invasive monitoring of cardiovascular haemodynamics, detection and localization of coronary arterial lesion.

The annotation of the sequences plays an important role in the taxonomic characteristics, medical research, phylogenetic studies, etc. Most of the existing classification techniques focused on some distinct group of organisms and considered the specific fragments of genome sequence, such as particular gene or RNAs. Here, we introduce a novel representation of genome sequence in a 2D coordinate space and extract features from the 2D representation. These features are used to classify a set of species into different taxonomy ranks. We experiment our proposed method, GenFooT, on nine different datasets of genomic sequences of various organisms. Our experimental results indicate improved classification performance of the proposed features with both Linear discriminant analysis and Logistic regression classifiers. This study demonstrates that our proposed alignment-free method, GenFooT, is fast, accurate, and can be applied to the large-scale genomics studies.

Specific learning disabilities are a major obstacle in early learning processes and are a growing issue in India. Children with specific learning disabilities lack in necessary skills of reading and writing and require a personalized intervention by a clinical expert. However, the low expert-to-populace proportion is a noteworthy obstacle in effectively treating the disorder in the fully human-guided therapeutic set-up. The emerging use of Android handsets and effective e-learning technologies in the modern era provides ways to reduce the physical distance between the clinician and the child. This paper proposes one of the many solutions to help children with dyslexia in their early learning. This paper proposes an Android game-based intervention program to teach reading at the word level. The main objective of this work is to lessen the dependency of experts by providing a unified platform to assist both experts and children. The games are designed based on the sight words training a widely used and accepted intervention strategy. This work shows the developed prototype of the proposed approach, and the reviews from subject matter experts on the prototype through the Mobile Application Rating Scale.

The recent works in tensor decomposition of convolutional neural networks have paid little attention to finetuning the decomposed models more effectively. We propose to improve the accuracy as well as the adversarial robustness of decomposed networks over existing noniterative methods by distilling knowledge from the computationally intensive undecomposed (teacher) model to the decomposed (student) model. Through a series of experiments, we demonstrate the effectiveness of knowledge distillation with different loss functions and compare it to the existing fine-tuning strategy of minimizing crossentropy loss with ground truth labels. Finally, we conclude that the student networks obtained by the proposed approach are superior not only in terms of accuracy but also adversarial robustness, which is often compromised in the existing methods.

Deep learning has proved to be highly efficient towards large scale, high dimensional data, rendering it to be an area of interest for researchers. Enormous volumes of satellite imagery enables us to utilise the benefits of a deep framework in the field of remote sensing. In this paper, we propose an unsupervised patch based learning method to cluster hybrid polarimetric SAR images. We extract small patches from the image data set, and train VGG16 model with batch normalization using an entropy based loss function. Initially, the patches are segmented into three classes, namely, surface, volume, and double-bounce, which are defined with reference to the SAR scattering characteristics. We further classify volume into dense vegetation, and agricultural areas. Mixed classes, mainly covering the areas which have settlements surrounded by tall trees, are also observed. This technique gives an average accuracy of 89.70%.

Coastline detection, and monitoring have different research challenges. Global warming, deforestation, and sea level rising have several adverse effects on coastlines such as, erosion, coastline recession, loss of biodiversity, etc. In the past, various techniques have been proposed to detect, and monitor coastlines. While detecting coastlines, most of the techniques do not consider removing inland water bodies. The objective of this work is to remove inland water bodies, in the process of detection of coastlines. For accomplishing this task, we propose to use difference of Coal Mine Index (CMI) and Normalized Difference Water Index (NDWI) to separate water bodies from other regions. Subsequently, inland water bodies are identified from this set by computing and analyzing their contours, and removed to provide coastlines. Overall, this work presents a novel automated method of coastline detection with the presence of inland water bodies having precision, and recall of 80%, and 82.35%, respectively.

Coal seam fire has various environmental, social, and economical adversities. In the past, coal seam fire regions are detected by studying the land surface temperature properties of satellite images. Yet, these techniques do not consider the spectral properties of the fire locations. Thus, various other high temperature regions are falsely detected as coal seam fire regions. The objective of this paper is to detect coal seam fire regions by analyzing the land surface temperature and eliminating the falsely detected high temperature regions. We propose a novel technique for detecting coal seam fire regions by using clay mineral ratio, which can differentiate other high temperature regions from coal seam fire regions.

Sentinel-2 provides a large volume of the multi-spectral multi-resolution dataset. Training deep convolutional architecture with such a dataset is still a challenging task in land cover classification due to the absence of ground truth. Also, the selection of appropriate deep architecture for handling the Sentinel-2 dataset is another challenging task. In this paper, we propose a convolutional neural network (CNN) architecture to extract the information from various combinations of bands in Sentinel-2 imagery. We use a loss function, proposed in an earlier work, to train our model in an unsupervised manner. Recent advances in deep learning allow for transferring knowledge from one data set to another. Thus, in our study, we aim at analyzing the “transfer learning” capabilities of our proposed network to land cover classification in Sentinel-2 images. Pre-trained weights of a deep neural architecture, which is trained with very high-resolution optical satellite imagery from Aviris is transferred to a network of almost similar architecture for processing Sentinel-2 data. We used Salinas, a publicly available hyper-spectral dataset, to train this architecture in a supervised fashion and, finally, use transfer learning and fine-tuning on the architecture handling Sentinel-2 data for clustering. Experiments show that bands of 60m resolution have a positive combining effect with bands of 20m resolution for segregating waterbody, urban settlement, and tree canopies distinctly.

Human gaze dynamics mainly concern about the sequence of the occurrence of three eye movements: fixations, saccades, and microsaccades. In this paper, we correlate them as three different states to velocities of eye movements. We build a state trajectory estimator based on ancestor sampling (ST EAS) model, which captures the features of the human temporal gaze pattern to identify the kind of visual stimuli. We used a gaze dataset of 72 viewers watching 60 video clips which are equally split into four visual categories. Uniformly sampled velocity vectors from the training set, are used to find the best suitable parameters of the proposed statistical model. Then, the optimized model is used for both gaze data classification and video retrieval on the test set. We observed 93.265% of classification accuracy and a mean reciprocal rank of 0.888 for video retrieval on the test set. Hence, this model can be used for viewer independent video indexing for providing viewers an easier way to navigate through the contents.

Human behavior can be analyzed using Eye tracker. Thus, it is used for revealing the cognitive processes for object identification. Cognitive process is the mental ability for identification of what our eyes see. Vision with 20/20 sometimes may not reveal the purpose. In this study, ambiguous images are taken to observe the cognitive process in participants. During the perception of an object, a participant uses goal-directed search for identifying various objects. Dense gaze coordinates provide the region of interests and are considered as the target regions for object identification in ambiguous images. These data are used to develop cognitive models for identification of ambiguous images. Features such as, eye fixation, pupil diameter, fixation durations, moments of inertia, and polar moments are used for developing the cognitive model. Three different feature selection methods along with six different classifiers are used for the task of classification. The selection of a subset of features using hypothesis testing performed well, compared to principal component analysis based dimensionality reduction method. This study could be used in detecting whether a participants is lying or not while perceiving an ambiguous image.

We present the Fitness Based Layer Rank Selection (FLRS) Algorithm for Accelerating Convolutional Neural Networks by CANDECOMP/PARAFAC (CP) Decompositions. FLRS selects the layers and corresponding ranks based on a parameter fitness factor. The advantage of the proposed FLRS algorithm is that it does not require retraining iteratively during rank selection. The experimental results show that VGG-16 Network can be replaced by an approximate network where the convolutional layers are replaced by a sequence of four convolutional layers with smaller kernels. The approximated network has less than one-fifth of the original model parameters and performs less than one-fifth of the total number of computations as compared to the original model with an accuracy drop of less than 1% across SVHN, CIFAR-10 and CALTECH-101 datasets.

In this paper, we propose to distinguish forest-cover in an unsupervised fashion by a combination of passive multi-spectral imagery and active hybrid polarized SAR data. At first, multi-spectral imagery (MSI) is used to separate general vegetation region (e.g., forest, mature grassland, and pre-harvest agricultural fields) from the imaged scene using spectral slopes based rules and support vector machine technique. Then, hybrid polarimetric SAR image of the same region (acquired with a common time stamp) is clustered into three scatter classes, namely, surface, volume, and dihedral, using Stokes parameters based m - δ decomposition. Forest cover is extracted by bi-labeled pixels of the study site that correspond to vegetation (in MSI) and volume scatter (in SAR), which forms a community level classification of forest region. Further, using Wishart derived mean-shift clustering technique, we segregate possible categories of forest clusters within the mapped forest region to obtain a sub-community level classification. Discernible spectral and scattering characteristics of remotely sensed images are explored in our work for identifying forest regions and their possible categories. The proposed method is automated by freeing the manual supervision in selecting seed pixels for training any machine learning technique.

Detection, and monitoring of surface mining region have various research aspects. Coal surface mining has severe social, ecological, environmental adverse effects. In the past, semisupervised and supervised clustering techniques have been used to detect such regions. Coal has lower reflectance values in short wave infrared I (SWIR-I) than short wave infrared II (SWIR-II). The proposed method presents a novel approach to detect coal mine regions without manual intervention using this cue. Clay mineral ratio is defined as a ratio of SWIR-I to SWIR-II. Here, unsupervised K-Means clustering has been used in a hierarchical fashion over a variant of clay mineral ratio to detect opencast coal mine regions in the Jharia coal field (JCF), India. The proposed method has average precision, and recall of 76.43%, and 62.75%, respectively.

We present a novel video summarization model to generate coherent video summaries capturing the most interesting parts of a video, using CNN and bidirectional LSTMs to generate deep features for frame representation and to model variable-range temporal sequences. Further, we introduce a parameterized loss function minimizing KL-divergence between GMMs to learn relative orders of frame importances. We conduct extensive evaluation on several benchmarks (TV-Sum, SumMe and YouTube) to demonstrate the effectiveness of our model, where our approach significantly outperforms state-of-the-art methods in several settings. Given the enormous growth in user-generated videos, video summarization has increasing importance in being able to navigate, browse, and search videos efficiently. Our research could see direct applications in tackling problems like detecting break - ins from surveillance videos, generating sporting event highlights, etc.

Filtering of images is required in various applications of image processing. Previously a few algorithms have been reported for filtering of images in the block discrete cosine transform (DCT) domain given its 2D finite impulse response (FIR). However, they assume that the response should be separable along rows and columns. In this paper, we propose a novel technique for implementing a non-separable 2D FIR filter in the block DCT domain. In this approach, we decompose a nonseparable 2D arbitrary FIR into a set of 2D separable arbitrary FIRs, and apply the computation of separable filters in the DCT domain. We have further used the sparsity of the DCT blocks and ranking of separable components in representing the nonseparable filter to reduce the computational cost allowing a graceful degradation of the quality of the filtered output.

Person Re-identification (Person re-id) is a crucial task as its application in visual surveillance and human-computer interaction is increasing day-by-day. In this work, we present a deep learning approach for video based person re-id problem. We use residual network (ResNet) along with LSTM for feature extraction. The extracted feature is passed through an attentive temporal pooling layer, which enables the feature extractor to be aware of the current input video sequences. In this way, inter dependency between two images can directly influence the computation of each other's feature representation. At last, we re-rank the result using k-reciprocal encoding method to mitigate the effect of false matching. Experiments conducted on iLIDS-VID and PRID 2011 datasets confirm that our model outperforms existing state-of-the-art video-based re-id methods.

Automatic segmentation of tumor in low dose scans like the Cone Beam Computed Tomography (CBCT) is quite challenging. We use a semi-automatic approach to segment tumor from non tumor using the classical level-set formulation. A pipeline of techniques, mainly involving gradient-based level-sets (GB) and Local Rank Transform (LRT) is used to achieve the tumor segmentation. Since CBCT images are prone to noise, the edge strength at the tumor and non-tumor boundary is very low. To improve the edge strength in the CBCT image, we propose to use the edges obtained from the LRT-attractor of the image. The gradient-based level-sets with LRT-attractor (GBLA) is a non-linear technique that helps in strengthening the latent tumor and non-tumor boundary. We compare the GBLA level-sets with the GB level-sets technique, and report our results on 307 volumes of 45 patients. It was found that average precision is improved by 10% when using GBLA.

Surface mining causes drastic land cover and land use changes and it has huge ecological, and environmental impacts. Restoration, reclamation, classification, and monitoring of surface mining region have several research aspects. In classification of surface mining, water bodies in mining regions are detected along with water bodies in non mining areas using various spectral information of satellite imagery. Water bodies in mining region have distinct characteristic. Detection of these distinct characteristics, and separation of mine, and non mine water body areas are the motivation of this paper. Normalized Difference Water Index (NDWI) is a widely practiced water detection technique which is used here. Detection of water bodies from NDWI may falsely detect few bare soil regions. Hence, in this work, bare soil regions are detected by Bare soil Index (BI) and further removed from the water map. The modified water map is analyzed by connected component computation and further mine, and non mine water bodies are separated by using the concept of clay mineral ratio.

Reliable localization is one of the most important parts of an MAV system. Localization in an indoor GPS-denied environment is a relatively difficult problem. Current vision based algorithms track optical features to calculate odometry. We present a novel localization method which can be applied in an environment having orthogonal sets of equally spaced lines to form a grid. With the help of a monocular camera and using the properties of the grid-lines below, the MAV is localized inside each sub-cell of the grid and consequently over the entire grid for a relative localization over the grid. We demonstrate the effectiveness of our system onboard a customized MAV platform. The experimental results show that our method provides accurate 5-DoF localization over grid lines and it can be performed in real-time.

Filtering and enhancement of images are two fundamental preprocessing operations, which are required in various applications of image processing. In this paper we review a few of our prior works in the block DCT domain.

In this study, the authors propose a new feedback‐channel‐free distributed video coding algorithm using local rank transform (LRT). The encoder computes LRT by considering selected neighbourhood pixels of Wyner–Ziv (WZ) frame. These LRT values are merged, and their positions are entropy coded and sent to the decoder. In addition, means of each block of WZ frame are also transmitted to assist motion estimation (ME). Using these measurements, the decoder generates side information (SI) by implementing ME and compensation in LRT domain. An iterative algorithm is executed on SI using LRT to reconstruct the WZ frame. Experimental results show that the coding efficiency of the authors’ codec is close to the efficiency of pixel domain distributed video coders based on low‐density parity check and accumulate or turbo codes, with less encoder and decoder complexity.

Phylogenetic supertrees synthesize a set of phylogenetic trees carrying overlapping taxa set, preferably with the consensus topologies of individual taxa subsets. Supertree construction is an NP-hard problem, and the methods based on decomposition and synthesis of fixed size subtree topologies (such as triplets or quartets) are the most popular. Time and space complexities of these methods, however, depend on the subtree size considered. Our earlier work proposed a couplet (taxa pair) based supertree method COSPEDTree, which produces slightly conservative (not fully resolved) supertrees. Here we propose its improved version COSPEDTree-II, which produces better resolved supertrees with lower number of missing branches, and incurs much lower running time.

Delivering medical care to newborn babies in their early days of life, involves complex mathematical calculation for feeding, intravenous fluid and electrolytes requirements. Manual calculation of this process is time consuming and potential source of medical error. This work proposes a standalone Android application for newborn care unit, which can run in any handheld Android device like mobile phones and helps health-care professionals to calculate certain parameters regarding the feed and intravenous fluid to be given to a newborn baby. The parameters are - total fluid intake, Glucose Infusion Rate, energy, protein, lipid amount, electrolytes, etc. Its logic is based on the medical guidelines for feed and fluid management of newborn babies. It maintains consistency in a large set of inter-related variables using an existential abstraction approach excluding the possibility of having wrong proportions of dextrose, protein, lipid or fluid volume by showing error and warning messages wherever needed, which acts as a safety measure to avoid medication errors. The objective of the work is to make the medical calculation process faster, safer and accurate. A prototype of the application is being tested in a Sick Newborn Care Unit (SNCU) in Kolkata,India for evaluation.

Rubber stamps often overlap with original text content of a document, and hence obscure the text regions very badly. Removal of these stamp regions becomes a necessity for successful conversion of such documents into electronic format. Stamp removal from a document becomes more difficult when they are in gray scale, or text and stamp are of the same color. In this paper, we propose a technique to remove such stamps from overlapped regions by identifying stamp regions and stamp pixels.

In the literature, various techniques for supervised/ semi-supervised classification of satellite imageries require manual selection of samples for each class. In this paper, we propose a spectral-slope based classification technique, which automates the process of initial labeling of a set of sample points. These are subsequently used in a supervised classifier as training samples and it performs the task of classification over all the pixels in the image. We demonstrate the effectiveness of our proposed classification technique in summarizing the changes in temporal image sets. For selecting the training samples from the satellite imageries, a set of rules is proposed by using the spectral-slope properties. We classify the land-cover into three classes, namely, water, vegetation, and vegetation-void, and validate the classification results using very high resolution satellite imagery. The approach has also been used in the analysis of images acquired by different sensors operating under similar wavelength ranges.

In this paper, we present a novel formulation of exemplar-based image inpainting as a metric labeling problem, and solve it through the simulated annealing algorithm. Due to their greedy nature, exemplar-based methods sometimes produce inpainted images, which are visually inconsistent. These methods are highly dependent upon the initialization. To solve these problems, we generate five images with a different initialization. A suitable mixture of these five images produces a good inpainted image. The cost function of the proposed metric labeling problem consists of three components, namely, neighbor cost, total variation cost, and structure cost. A linear combination among these components is used to maintain better visual consistency in the inpainted region having smooth transition from the bordering regions of the source image. We use a quality measure to this end. Our experiments on a wide variety of images demonstrate that the proposed technique produces better inpainting images as compared with some other state-of-the-art techniques.

Text and graphics separation is an important task in the field of document image processing. This work aims at detecting graphics such as logos and stamps in a scanned document image. A novel spectral filtering based text-graphics separation algorithm (SFTGS) is presented here. The property of text that it is the major source of high spatial frequency components in a document image, is exploited in this algorithm. Accordingly high frequency filtering is used to separate the text symbols. This is followed by a segmentation process for delineating residual text and the graphics. The main advantage of SFTGS is that it works in a single pass, and can discriminate graphics and text without supervised training. Subsequently, the graphics segments are further categorized into two different classes, namely logos and stamps. In this case, we assume that these are the two classes of graphical objects present in the documents. The technique is evaluated using publicly available document dataset consisting of graphics as stamps and logos. The result is compared with existing approaches reported in the literature, and it is found that the proposed method performs superior to them. An overall performance of 89.1% recall and 96.9% precision is obtained for SFTGS.

Occlusion is one of the most challenging problems in many video processing applications such as surveillance, gait recognition, activity recognition and so on. Attempts have been made to develop algorithms for handling occlusion and evaluate their performance on various datasets. However, these studies are subjective in nature and the datasets are hardly characterised in terms of the level of occlusion, thereby precluding any form of quantitative comparison of performance. This shows a compelling need to design an explicit, unambiguous and quantitative model, which should be able to objectively represent occlusion in a video. This study proposes an occlusion model based on the position and pose uncertainties of the moving subjects in a video. The proposed occlusion model is able to characterise the level of occlusion present in a video. It is also employed to synthetically generate occlusion for walking sequences, thus providing a direction for controlled dataset generation against which human identification algorithms can be tested. Given an input video with a subject moving without any occlusion, a particle swarm optimisation‐based parameter estimation methodology is presented that generates the desired level of occlusion. The proposed approaches have been tested on the TUM‐IITKGP and PETS2010 datasets. Finally, as an application, the occlusion model has been used to generate an occluded gait datasets and the performances of different gait recognition algorithms have been compared under varying levels of occlusion.

In this study, the authors present a fully automated frontal (i.e. employing front and back views only) gait recognition approach using the depth information captured by multiple Kinect RGB‐D cameras. Limited depth sensing range restricts each of these Kinects to record only a part of a complete gait cycle of a walking subject. Hence, information from more than one Kinect is fused together to examine which features of a gait cycle can be conveniently extracted from the sequences captured independently by these cameras. To achieve this, it is imperative that the same subject be re‐identified as he moves from the field of view of one camera to another. The authors use a set of soft‐biometric features computed from the skeleton stream provided by Kinect software development kit) for doing automatic re‐identification. To enable such information fusion and also to handle missing components even after re‐identification, features are extracted at the granularity of small fractions of a gait cycle. Experiments carried out on a data set with gait videos captured by Kinects respectively from the back and front views show promising results.

From a set of phylogenetic trees with overlapping taxa set, a supertree exhibits evolutionary relationships among all input taxa. The key is to resolve the contradictory relationships with respect to input trees, between individual taxa subsets. Formulation of this NP hard problem employs either local search heuristics to reduce tree search space, or resolves the conflicts with respect to fixed or varying size subtree level decompositions. Different approximation techniques produce supertrees with considerable performance variations. Moreover, the majority of the algorithms involve high computational complexity, thus not suitable for use on large biological data sets. Current study presents COSPEDTree, a novel method for supertree construction. The technique resolves source tree conflicts by analyzing couplet (taxa pair) relationships for each source trees. Subsequently, individual taxa pairs are resolved with a single relation. To prioritize the consensus relations among individual taxa pairs for resolving them, greedy scoring is employed to assign higher score values for the consensus relations among a taxa pair. Selected set of relations resolving individual taxa pairs is subsequently used to construct a directed acyclic graph (DAG). Vertices of DAG represents a taxa subset inferred from the same speciation event. Thus, COSPEDTree can generate non-binary supertrees as well. Depth first traversal on this DAG yields final supertree. According to the performance metrics on branch dissimilarities (such as FP, FN and RF), COSPEDTree produces mostly conservative, well resolved supertrees. Specifically, RF metrics are mostly lower compared to the reference approaches, and FP values are lower apart from only strictly conservative (or veto) approaches. COSPEDTree has worst case time and space complexities of cubic and quadratic order, respectively, better or comparable to the reference approaches. Such high performance and low computational costs enable COSPEDTree to be applied on large scale biological data sets.

Power Integrity is maintained in a high speed system by designing an efficient decoupling network. This paper provides a generic formulation for decoupling capacitor selection and placement problem which is solved by mixed-integer programming. A real-world example is presented for the same. The minimum number of capacitors that could achieve the target impedance over the desired frequency range are found along with their optimal locations. In order to solve an industrial problem, the s-parameters data of power plane geometry and capacitors are used for the accurate analysis including bulk capacitors and VRM.

Frontal gait recognition using partial cycle information has not received significant attention to date in spite of its many potential applications. In this paper, we propose a hierarchical classification strategy that combines front and back view features captured by RGB-D (Red Green Blue - Depth) cameras. Airport security check points are considered as a typical application scenario, where two depth cameras mounted on top of a metal detector gate positioned beyond a yellow line, respectively, record front and back views of a subject as he goes through the check-in process. Due to the short distance of the surveillance zone between the yellow line and point of exit, it is often not possible to capture a full gait cycle independently from the front view or back view. An initial stage of anthropometric feature-based classification followed by motion feature extraction from the front view is used to restrict the potential set of matched subjects. A final classification is then applied on this reduced set of subjects using depth features extracted from the back view. The method is computationally efficient with a much higher rate of accuracy compared with existing gait recognition approaches.

Skew correction of a scanned document page is an important preprocessing step in document image analysis. We propose here a fast and robust skew estimation algorithm based on rank analysis in Farey sequence. Our target document class comprises two major Indian scripts with headlines, namely Devnagari and Bangla. At the beginning, straight edge segments from the edge map of the document page are detected by our algorithm using properties of digital straightness. Straight edges derived in this manner are binned by Farey ranks in correspondence with their slopes. The principal bin, identified from these bins using the strength of accumulated edge points, represents the principal direction along the direction of headlines, from which the gross skew angle is estimated. A fast refinement algorithm is then applied with a finer tuning of Farey ranks, to detect the skew up to the desired level of precision. The algorithm has been tested on a diverse set of document images, containing Bangla and Devnagari scripts. Experimental results are quite encouraging in terms of accuracy, sensitivity to non-textual objects, effectiveness in dealing with unrestricted layouts, and computational efficiency.

Swarm intelligence is applied to a module of high speed system design problem. To maintain power integrity in a high speed system, an effective methodology for suppressing the cavity-mode anti-resonances' peaks is presented. The optimal values and the optimal positions of the decoupling capacitors are found using three different swarm intelligence methods - particle swarm optimization, cuckoo search method and firefly algorithm. Optimum values and locations of decoupling capacitors are obtained, by which anti-resonances' peaks of loaded board are minimized.

This paper proposes a method for alignment of human brain magnetic resonance (MR) image sequences in the brain based on a 3D human brain model (triangulated mesh). The brain model is composed of four components, namely, cerebrum, cerebellum, brain stem and pituitary gland which are represented by four different colors. Synthesized image sequences (cross-sections) are extracted from the model at regular intervals for sagittal and coronal views as done in MR imaging. The cerebellums are segmented from the sequence of MR images by using the method of active contouring and their sizes are determined. The areas of the cerebellums are computed from the cross-sections using the color information. To obtain the optimal synthesized cross-section sequence corresponding to the series of MR images, an efficient dynamic programming based computational technique has been developed that uses the normalized sizes of cerebellum in both the MR image sequences and the cross-sections.

Information Systems are needed for modernization of ICUs to deliver better health care services. EHR systems can improve the work flow management in health care delivery. This work proposes a secure web enabled system based on a multi-tier architecture for carrying out routine and special operations of Neonatal Intensive Care Unit (NICU). The system adopts a service oriented approach for execution of various tasks that are performed for managing NICU activities. It also facilitates decision support systems for a number of critical tasks of NICU. A prototype of the system has been installed in the neonatology department of SSKM Hospital, Kolkata, India and the staff of the hospital including doctors, nurses, laboratory personals and technicians are using it in a regular manner.

In this work, we present a simple expression relating the input, output and the filter responses in the block DCT space by isolating contributions of each neighboring block. The proposed technique considers only filters with non-causal symmetric response. The method provides the exact computation of the linear convolution in the compressed domain. Though it is computationally less efficient than existing techniques, the technique is simpler to implement and is suitable for parallel implementation. Further, it also provides a novel framework for computing with both sparse and non-sparse data at the same time and thus reducing the computational requirement substantially with significant improvement in the quality of the output.

This paper considers different algorithms for solving the color constancy problem in the block DCT domain. New algorithms in this domain have been proposed by exploiting the advantage of smaller data size and the better chromatic separation of Y-Cb-Cr color space (as used in the JPEG standard) against the R-G-B color space of the spatial domain. These make them faster and less memory intensive compared to their equivalent ones in the spatial domain without any significant degradation of the quality of the results.

Transcoding enables the transformation of multimedia content to adapt to a diverse nature of client/user requirements. In this paper, we propose a technique for transcoding wavelet coefficients to block DCT coefficients in the transform domain. In the first step, the wavelet coefficients are transformed into upsampled block DCT coefficients. Subsequently these transformed coefficients are synthesized by filtering in the block DCT space. To reduce the transcoding complexity, we perform upsampling and filtering operations in a single combined step. The proposed transcoding approach restricts all operations of the synthesis process in the block DCT space. The proposed approach achieves the same quality of reconstruction (Except rounding errors in processing) as that of spatial domain technique with reduced complexity for a given wavelet synthesis filter bank.

This paper presents a new efficient technique for color enhancement in the compressed domain. In the proposed technique the enhancement computation is performed in three steps, namely, adjustment of dynamic range of brightness values, preservation of local contrast and preservation of color with all computations involving scaling of the DCT coefficients. Further special care has been taken for reducing the blocking artifacts. The proposed algorithm provide better performances compared to some of the existing techniques in the compressed domain.

This paper presents a new technique for color enhancement in the compressed domain. The proposed technique is simple but more effective than some of the existing techniques reported earlier. The novelty lies in this case in its treatment of the chromatic components, while previous techniques treated only the luminance component. The results of all previous techniques along with that of the proposed one are compared with respect to those obtained by applying a spatial domain color enhancement technique that appears to provide very good enhancement. The proposed technique, computationally more efficient than the spatial domain based method, is found to provide better enhancement compared to other compressed domain based approaches.

In this paper, we propose a hierarchical state-based model for representing an echocardiogram video. It captures the semantics of video segments from dynamic characteristics of objects present in each segment. Our objective is to provide an effective method for segmenting an echo video into view, state, and substate levels. This is motivated by the need for building efficient indexing tools to support better content management. The modeling is done using four different views, namely, short axis, long axis, apical four chamber, and apical two chamber. For view classification, an artificial neural network is trained with the histogram of a region of interest of each video frame. Object states are detected with the help of synthetic M-mode images. In contrast to traditional single M-mode, we present a novel approach named sweep M-mode for state detection. We also introduce radial M-mode for substate identification from color flow Doppler 2-D imaging. The video model described here represents the semantics of video segments using first-order predicates. Suitable operators have been defined for querying the segments. We have carried out experiments on 20 echo videos and compared the results with manual annotation done by two experts. View classification accuracy is 97.19%. Misclassification error of the state detection stage is less than 13%, which is within acceptable range since only frames at the state boundaries are found to be misclassified.

In this paper, we propose a graph-based approach for detecting and tracking multiple players in broadcast soccer videos. In the first stage, the position of the players in each frame is determined by removing the non player regions. The remaining pixels are then grouped using a region growing algorithm to identify probable player candidates. A directed weighted graph is constructed, where probable player candidates correspond to the nodes of the graph while each edge links candidates in a frame with the candidates in next two consecutive frames. Finally, dynamic programming is applied to find the trajectory of each player. Experiments with several sequences from broadcasted videos of international soccer matches indicate that the proposed approach is able to track the players reasonably well even under varied illumination and ground conditions.

In many multi-media applications DCT based digital video coding standards like MPEG are widely used. In certain applications like video database browsing, picture in picture etc., the video is required to be downscaled before transmitting it over the network. The naive spatial domain approach for video downscaling requires video to be fully decompressed. It requires huge computation time. The computation time greatly be reduced if all the computations are performed in the DCT domain. In this paper, we propose a fast DCT domain based video downscaling system. To reduce computational requirement, we have proposed efficient algorithms for inverse motion compensation. The sparseness of the DCT blocks are also considered for further reduction of computations. We found that our proposed system is 36 times faster than the spatial domain method and PSNR wise produces better quality downscaled video. Index Terms– Discrete cosine transform (DCT), MPEG, video downscaling, inverse motion compensation, motion-vector refinement.

We propose a fast algorithm to down-size video frames, by arbitrary factor, directly in DCT domain. The existing methods treat each 8×8 block as a fundamental unit, therefore, involve high cost of reconstructing down-sized frames. We demonstrate that a basic operation in reconstructing 16×16 macroblock, as a whole, can be represented as multiplication by fixed matrices and the computations can be greatly simplified through decomposition of DCT/IDCT matrix operations. Experimental results using cascaded DCT domain transcoder (CDDT) show substantial reduction in cost of reconstructing down-sized frames at quality similar to the existing methods.

We present a new algorithm for resizing video frames in the discrete cosine transform (DCT) space. We demonstrate that a frame resizing operation can be represented as multiplication by fixed matrices and propose a computation scheme which is applicable to any DCT-based compression method. The proposed approach is general enough to accommodate resizing operations with arbitrary factors conforming to the syntax of 16times16 macroblocks. The approach is shown to possess significant computational gain over the faster known state of the art algorithms while achieving similar picture quality

In this paper, we present a graph-theoretic interpretation of convergence of fractal encoding based on partial iterated function system (PIFS). First we have considered a special circumstance, where no spatial contraction has been allowed in the encoding process. The concept leads to the development of a linear time fast decoding algorithm from the compressed image. This concept is extended for the general scheme of fractal compression allowing spatial contraction (on averaging) from larger domains to smaller ranges. A linear time fast decoding algorithm is also proposed in this situation, which produces a decoded image very close to the result obtained by an ordinary iterative decompression algorithm.

Representation schemes play an important role in the fields of Computer Vision, Graphics, Image Processing, CAD/CAM etc. Various representation schemes have been discussed in the literature for both 2D and 3D. In this paper, we are presenting a scheme of representation using the concept of octagonal distances. They are called Medial Circle Representation (MCR) and Medial Sphere Representation (MSR) in 2D and 3D, respectively. Storage requirement, computational complexity, merits and demerits of the representation schemes are discussed.

This paper describes a system which obtains a structural scene description of 3-D objects from range images. The system uses a hierarchical approach to obtain higher level primitives from lower level ones. Instead of a detailed mathematical approach, qualitative reasoning by rule based deduction is used to obtain the scene description. The rule bases are also hierarchical and several special control strategies like rule pruning, windowing (or zoning) and fact inhibition are used to considerably improve the speed of the system. Experimental results and performance of the system on actual range images are presented.