Department of Computer Science & Engineering

&ldquo;IIT Jodhpur is committed to impart quality education to facilitate the creation of qualified, competent and responsible human resources to meet the emerging technological challenges of the world. It is working towards creating an equitable learning platform that is positive, accessible and effective in the face of the modern societal and scientific advancements.

Flexibility in curriculum, in all its academic programs, encompassing elements of teaching, research, creativity, innovation and entrepreneurship is the most unique feature of IIT Jodhpur. The Department of Science and Technology (DST), under the National Mission on Cyber Physical Systems, has identified IIT Jodhpur as a Technology &amp; Innovation Hub in the area of Augmented Reality and Virtual Reality. The institute has also been nominated as the coordinator for the Jodhpur City and Knowledge Cluster by the Office of Principal Scientific Adviser, Government of India.

The sprawling 852-acre IIT Jodhpur campus is a majestic tribute to the deep-rooted diverse cultural heritage of the region. The campus is an architectural masterpiece, which is slated to become an international exemplar of sustainability with net-zero energy, and advanced water and waste management strategies.&rdquo;

IIT Jodhpur

Deep LSTM is an ideal candidate for text recognition. However text recognition involves some initial image processing steps like segmentation of lines and words which can induce error to the recognition system. Without segmentation, learning very long range context is difficult and becomes computationally intractable. Therefore, alternative soft decisions are needed at the pre-processing level. This paper proposes a hybrid text recognizer using a deep recurrent neural network with multiple layers of abstraction and long range context along with a language model to verify the performance of the deep neural network. In this paper we construct a multi-hypotheses tree architecture with candidate segments of line sequences from different segmentation algorithms at its different branches. The deep neural network is trained on perfectly segmented data and tests each of the candidate segments, generating unicode sequences. In the verification step, these unicode sequences are validated using a sub-string match with the language model and best first search is used to find the best possible combination of alternative hypothesis from the tree structure. Thus the verification framework using language models eliminates wrong segmentation outputs and filters recognition errors. © 2015 IEEE.

World University root node

Proceedings of the International Conference on Document Analysis and Recognition, ICDAR

A hypothesize-and-verify framework for text recognition using deep recurrent neural networks

With the rapid growth of broadcast systems and ease of accessing internet services, lots of information is available and accessible on the web. The information available in multimedia documents may have different context and content. Since, interpretation of multimedia content cannot be free of context so tagging on the basis on context is indispensable for dealing with this problem. Tagging plays an important role in retrieving multimedia data as now-a-days most of the videos are retrieved based on text describing them and not by the actual context embodied in them. So, in this paper we have proposed a scheme for tagging multimedia data based on the contents and context as identified from web-based resources. The hierarchical LDA (hLDA) is used to model the context information while Correspondence-LDA (Corr-LDA) is used to model the content information of multimedia data. Finally, multimedia data is tagged with the relevant contents and context information on the basis of Context-Matching Algorithm. These tags can then be used by search engines for increasing precision and recall of multimedia search results. © Springer International Publishing Switzerland 2015.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Context-based semantic tagging of multimedia data

In this paper we have presented a technique for enhancement and retrieval of historic inscription images. Inscription images in general have no distinction between the text layer and background layer due to absence of color difference and possess highly correlated signals and noise; pertaining to which retrieval of such images using search based on feature matching returns inaccurate results. Hence, there is a need to first enhance the readability and then binarize the images to create a digital database for retrieval. Our technique provides a suitable method for the same, by separating the text layer from the non-text layer using the proposed cumulants based Blind Source Extraction(BSE) method, and store them in a digital library with their corresponding historic information. These images are retrieved from database using image search based on Bag-of-Words(BoW) method. © Springer International Publishing Switzerland 2015.

Enhancement and retrieval of historic inscription images

Biometrics, an integral component of Identity Science, is widely used in several large-scale-county-wide projects to provide a meaningful way of recognizing individuals. Among existing modalities, ocular biometric traits such as iris, periocular, retina, and eye movement have received significant attention in the recent past. Iris recognition is used in Unique Identification Authority of India's Aadhaar Program and the United Arab Emirate's border security programs, whereas the periocular recognition is used to augment the performance of face or iris when only ocular region is present in the image. This paper reviews the research progression in these modalities. The paper discusses existing algorithms and the limitations of each of the biometric traits and information fusion approaches which combine ocular modalities with other modalities. We also propose a path forward to advance the research on ocular recognition by (i) improving the sensing technology, (ii) heterogeneous recognition for addressing interoperability, (iii) utilizing advanced machine learning algorithms for better representation and classification, (iv) developing algorithms for ocular recognition at a distance, (v) using multimodal ocular biometrics for recognition, and (vi) encouraging benchmarking standards and open-source software development. © 2015 Elsevier B.V. All rights reserved.

Information Fusion

Ocular biometrics: A survey of modalities and fusion approaches

Research in face recognition under constrained environment has achieved an acceptable level of performance. However, there is a significant scope for improving face recognition capabilities in unconstrained environment including surveillance videos. Such videos are likely to record multiple people within the field of view. Face recognition in such a setting poses a set of challenges including unreliable face detection, multiple subjects performing different actions, low resolution, and sensor interoperability. In general, existing video face databases contain one subject in a video sequence. However, real world video sequences are more challenging and generally contain more than one person in a video. Therefore, in this paper, we provide an annotated crowd video face (ACVF-2014) database, along with face landmark information to encourage research in this important problem. The ACVF-2014 dataset contains 201 videos of 133 subjects where each video contains multiple subjects. We provide two distinct use-case scenarios, define their experimental protocols, and report baseline verification results using OpenBR and FaceVACS. The results show that both the baseline results do not yield more than 0.16 genuine accept rate @ 0.01 false accept rate. A software package is also developed to help researchers evaluate their systems using the defined protocols. © 2015 IEEE.

Proceedings of 2015 International Conference on Biometrics, ICB 2015

Annotated crowd video face database

Medium bit rate hybrid speech coding schemes have gained much interest in the recent years and many of them have been standardized for various applications. This work characterizes a speech codec in a Compressive Sensing (CS) framework. We mainly demonstrate two aspects 1) Simultaneous compression and de-noising of speech by CS 2) Appropriate quantization of CS measurements to design medium bit-rate codec. The proposed scheme renders better quality speech compared to CELP, the widely used hybrid coding scheme, at the same bit rates. The CS speech codec has the added advantage of inherent noise suppression and easy scalability, without complex parameter extractions and voice activity detections. © 2015 IEEE.

2015 IEEE International Conference on Signal Processing, Informatics, Communication and Energy Systems, SPICES 2015

Speech coding and enhancement using quantized compressive sensing measurements

Phase estimation plays an important role in various signal processing areas like Radar, Sonar, power systems, speech analysis, communications and many others. The phase of the analytic form of the non stationary signals can be used to find instantaneous frequency. This paper addresses the problem of simultaneous phase and instantaneous frequency estimation from polynomial phase signals embedded in Gaussian noise. Here we have introduced the modified signal tracking approach which is then realized using unscented Kalman filter. The state space model is derived using Taylor series expansion of the phase of polynomial phase signal as process model while Polar to Cartesian conversion as measurement model. Proposed method, compared with state-of-the-art, performs better for signals with higher order polynomial phase variations at lower Signal-to-Noise-Ratio (0-5dB). We also present the simulation results for phase estimation. © 2015 IEEE.

Signal tracking approach for simultaneous estimation of phase and instantaneous frequency

A novel texture based color image enhancement methodology that focuses on an automated way of target image generation is proposed here. The images in the database with highest histogram correlation with input image are identified for extracting different features. Target image is obtained by fusing images selected based on minimum Euclidean distance between extracted features. The proposed method is a simple color image enhancement methodology where the range (the gamut) of the R, G, and B channels is optimally preserved. We derive a new quantitative validation approach to identify visibility loss problem that may occur during enhancement. The maximum possible contrast enhancement is achieved by stretching the intervals of the color levels to the maximum possible extent using a sigmoid function. The proposed method has been compared with other state-of-the-art algorithms reported in the literature. © 2015 The Authors.

Fulltext

Procedia Computer Science

A novel texture based automated histogram specification for color image enhancement using image fusion

Directionality analysis of time-series, recorded from task-activated regions-of-interest (ROIs) during functional Magnetic Resonance Imaging (fMRI), has helped in gaining insights of complex human behavior and human brain functioning. The most widely used standard method of Granger Causality for evaluating directionality employ linear regression modeling of temporal processes. Such a parameter-driven approach rests on various underlying assumptions about the data. The short-comings can arise when misleading conclusions are reached after exploration of data for which the assumptions are getting violated. In this study, we assess assumptions of Multivariate Autoregressive (MAR) framework which is employed for evaluating directionality among fMRI time-series recorded during a Sensory-Motor (SM) task. The fMRI time-series here is an averaged time-series from a user-defined ROI of multiple voxels. The 'aim' is to establish a step-by-step procedure using statistical methods in conjunction with graphical methods to seek the validity of MAR models, specifically in the context of directionality analysis of fMRI data which has not been done previously to the best of our knowledge. Here, in our case of SM task (block design paradigm) there is violation of assumptions, indicating the inadequacy of MAR models to find directional interactions among different task-activated regions of brain. © 2015 IEEE.