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

Machine learning methods are used today mostly for recognition problems. Convolutional Neural Networks (CNN) have time and again proved successful for many image processing tasks primarily for their architecture. In this paper we propose to apply CNN to small data sets like for example, personal photo albums or other similar environs where the size of training dataset is a limitation, within the framework of a proposed hybrid CNN-AIS model. We use Artificial Immune System Principles to enhance the small size of training data set. A layer of Clonal Selection is added to the local filtering and max pooling of CNN Architecture. The proposed Architecture is evaluated using the standard MNIST dataset by limiting the data size and also with a small personal data sample belonging to two different classes. Experimental results show that the proposed hybrid CNN-AIS based recognition engine works well when the size of training data is limited in size. © Springer International Publishing Switzerland 2015.

World University root node

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

A novel hybrid CNN-AIS visual pattern recognition engine

This paper proposes a novel approach for finding the orientation of the tilted pellet using a single view of the pellet. This is important for the automation of pick and place problem. A texture less isolated cylindrical object is used for the pick and place operation. In this approach the isolated pellets are segmented from the background, following which, its contour is estimated. Then based on the assumption that length and diameter of the pellet remains constant, the mathematical formulation of the pose estimation of the pellet is given, and using which the orientation of the tilted pellet is computed using just single view. This is then experimentally verified for the different orientations of the pellet and the results are within the acceptable levels of accuracy A brief overview of the error estimation has been included. This error in orientation of the pellet is due to the variance in the actual height of the pellet. Error Jacobian for the above inaccuracy was calculated, and was found to be within required limits. © 2015 ACM.

ACM International Conference Proceeding Series

Pose estimation of a tilted pellet using single view from robot mounted camera

Script based features are highly discriminative for text segmentation and recognition. Thus they are widely used in Optical Character Recognition(OCR) problems. But usage of script dependent features restricts the adaptation of such architectures directly for another script. With script independent systems, this problem can be solved to a certain extent for monolingual documents. But the problem aggravates in case of multilingual documents as it is very difficult for a single classifier to learn many scripts. Generally a script identification module identifies text segments and accordingly the script-dependent classifier is selected. This paper presents a unified framework of language model and multiple preprocessing hypotheses for word recognition from bilingual document images. Prior to text recognition, preprocessing steps such as binarization and segmentation are required for ease of recognition. But these steps induce huge combinatorial error propagating to final recognition accuracy. In this paper we use multiple preprocessing routines as alternate hypotheses and use a language model to verify each alternative and choose the best recognized sequence. We test this architecture for word recognition of Kannada-English and Telugu-English bilingual documents and achieved better recognition rates than single methods using same classifier. © 2015 IEEE.

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

OCR for bilingual documents using language modeling

In this paper, a parametric texture based video coding scheme is proposed for multi-view videos. The proposed scheme exploits the perceptual redundancy of the textured region in a statistical manner which is ignored in the multi-view video coding scheme of H.264/MVC. This statistical nature of texture is independent of the 3D scene structure. The set of most similar texture macroblocks over different frames is termed as a motion thread. In the proposed scheme, each P slice/frame of multi-view video is categorized into non-texture blocks which are encoded by H.264/MVC and texture blocks which are encoded by our novel approach. These texture based motion threads are encoded with the help of Spatio-Temporal Autoregressive (STAR) model. These motion threads covers the temporal and interview movement of macroblocks maintaining consistency while STAR model exploits the temporal and inter-view perceptual redundancy in neighborhood. The proposed scheme achieves average gain as high as 4.9 dB PSNR over H.264/MVC for given multi-view videos. © 2015 IEEE.

ICAPR 2015 - 2015 8th International Conference on Advances in Pattern Recognition

Parametric coding of texture in multi-view videos for 3DTV

The process of 2D to 3D involves depth map estimation of input image and stereo pair generation using that depth image. Fully automatic techniques are not able to give very high quality output and methods which involve human operators are costly and time consuming. We propose a semiautomatic technique which incorporates monocular depth cues if these are not present in the input image. We propose a method which first includes defocusing factor in the input image semi-automatically and then estimates depth from focus cue. We also used motion parallax as another monocular depth cue and evaluate the performance of final depth map. Gradient Magnitude Similarity Deviation which is a highly efficient perceptual quality index is used for the performance evaluation of final depth map. GMSD score is lowest in case of final depth map which shows highest quality. Finally, stereo pair is generated using original image and its depth map. Copyright 2015 ACM.

Stereo pair generation by introducing bokeh effect in 2D images

The depth cues which are also called monocular cues from single still image are more versatile while depth cues of multiple images gives more accurate depth extraction. Machine learning is a promising and new research direction for this type of conversion in today scenario. In our paper, a fast automatic 2D to 3D conversion technique is proposed which utilizes a fixed point learning framework for the accurate estimation of depth maps of query images using model trained from a training database of 2D color and depth images. The depth maps obtained from monocular and motion depth cues of input images/video and ground truth depths are used in training database for the fixed point iteration. The results produces with fixed point model are more accurate and reliable than MRF fusion of both types of depth cues. The stereo pairs are generated then using input video frames and their corresponding depth maps obtained from fixed point learning framework. These stereo pairs are put together to get the final 3D video which can be displayed on any 3DTV and seen using 3D glasses. © Springer International Publishing Switzerland 2015.

Fixed point learning based 3D conversion of 2D videos

Body weight variations are an integral part of a person's aging process. However, the lack of association between the age and the weight of an individual makes it challenging to model these variations for automatic face recognition. In this paper, we propose a regularizer-based approach to learn weight invariant facial representations using two different deep learning architectures, namely, sparse-stacked denoising autoencoders and deep Boltzmann machines.We incorporate a body-weight aware regularization parameter in the loss function of these architectures to help learn weight-aware features. The experiments performed on the extended WIT database show that the introduction of weight aware regularization improves the identification accuracy of the architectures both with and without dropout. © 2015 IEEE.

Fulltext

IEEE Access

Regularized deep learning for face recognition with weight variations

Recent developments in three-dimensional sensing devices has led to the proposal of a number of biometric modalities for non-critical scenarios. Leap Motion device has received attention from Vision and Biometrics community due to its high precision tracking. In this research, we propose Leap Password; a novel approach for biometric authentication. The Leap Password consists of a string of successive gestures performed by the user during which physiological as well as behavioral information is captured. The Conditional Mutual Information Maximization algorithm selects the optimal feature set from the extracted information. Match-score fusion is performed to reconcile information from multiple classifiers. Experiments are performed on the Leap Password Dataset, which consists of over 1700 samples obtained from 150 subjects. An accuracy of over 81% is achieved, which shows the effectiveness of the proposed approach. © 2015 IEEE.

2015 IEEE 7th International Conference on Biometrics Theory, Applications and Systems, BTAS 2015

A Leap Password based verification system

Large-scale fingerprint recognition involves capturing ridge patterns at different time intervals using various methods, such as live-scan and paper-ink approaches, introducing intraclass variations in the fingerprint. The performance of existing algorithms is significantly affected when fingerprints are captured with diverse acquisition settings such as multisession, multispectral, multiresolution, with slap, and with latent fingerprints. One of the primary challenges in developing a generic and robust fingerprint matching algorithm is the limited availability of large data sets that capture such intraclass diversity. In this paper, we present the multisensor optical and latent fingerprint database of more than 19000 fingerprint images with different intraclass variations during fingerprint capture. We also showcase the baseline results of various matching experiments on this database. The database is aimed to drive research in building robust algorithms toward solving the problem of latent fingerprint matching and handling intraclass variations in fingerprint capture. Some potential applications for this database are identified and the research challenges that can be addressed using this database are also discussed. © 2013 IEEE.

Multisensor Optical and Latent Fingerprint Database

Latent fingerprints are lifted from multiple types of surfaces, which vary in material type, texture, color, and shape. These differences in the surfaces introduce significant intra-class variations in the lifted prints such as availability of partial print, background noise, and poor ridge structure quality. Due to these observed variations, the overall quality and the matching performance of latent fingerprints vary with respect to surface properties. Thus, characterizing the performance of latent fingerprints according to the surfaces they are lifted from is an important research problem that needs attention. In this research, we create a novel multi-surface latent fingerprint database and make it publicly available for the research community. The database consists of 551 latent fingerprints from 51 subjects lifted from eight different surfaces. Using existing algorithms, we characterize the quality of latent fingerprints and compute the matching performance to analyze the effect of different surfaces. © 2015 IEEE.