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

This paper presents a novel methodology for tracking a single moving object in a video sequence applying the concept of rough set theory. The novelty of this technique is that it does not consider any prior information about the video sequence unlike many existing techniques. The first target model is constructed using the median filtering based foreground detection technique and after that the target is reconstructed in every frame according to the rough set based feature reduction concept incorporating a measure of indiscernibility instead of indiscernibility matrix. The area of interest is initially defined roughly in every frame based on the object shift in the previous frames, and after reduction of redundant features the object is tracked. The measure of indiscernibility of a feature is defined based on its degree of belonging (DoB) to the target. Three quantitative indices based on rough sets, feature similarity and Bhattacharya distance are proposed to evaluate the performance of tracking and detect the mis-tracked frames in the process of tracking to make those corrected. Unlike many existing measures, the proposed ones do not require to know the ground truth or trajectory of the video sequence. Extensive experimental results are given to demonstrate the effectiveness of the method. Comparative performance is demonstrated both visually and quantitatively.

World University root node

Fundamenta Informaticae

Unsupervised tracking, roughness and quantitative indices

A new spatio-temporal segmentation approach for moving object(s) detection and tracking from a video sequence is described. Spatial segmentation is carried out using rough entropy maximization, where we use the quad-tree decomposition, resulting in unequal image granulation which is closer to natural granulation. A three point estimation based on Beta Distribution is formulated for background estimation during temporal segmentation. Reconstruction and tracking of the object in the target frame is performed after combining the two segmentation outputs using its color and shift information. The algorithm is more robust to noise and gradual illumination change, because their presence is less likely to affect both its spatial and temporal segments inside the search window. The proposed methods for spatial and temporal segmentation are seen to be superior to several related methods. The accuracy of reconstruction has been significantly high. © 2012 Elsevier B.V. All rights reserved.

Applied Soft Computing Journal

Granulation, rough entropy and spatiotemporal moving object detection

We present an approach for the text-to-image retrieval problem based on textual content present in images. Given the recent developments in understanding text in images, an appealing approach to address this problem is to localize and recognize the text, and then query the database, as in a text retrieval problem. We show that such an approach, despite being based on state-of-the-art methods, is insufficient, and propose a method, where we do not rely on an exact localization and recognition pipeline. We take a query-driven search approach, where we find approximate locations of characters in the text query, and then impose spatial constraints to generate a ranked list of images in the database. The retrieval performance is evaluated on public scene text datasets as well as three large datasets, namely IIIT scene text retrieval, Sports-10K and TV series-1M, we introduce. © 2013 IEEE.

Proceedings of the IEEE International Conference on Computer Vision

Image retrieval using textual cues

We present a novel method for content-based recommendation of media-rich commodities using probabilistic multimedia ontology. The ontology encodes subjective knowledge of experts that enables interpretation of media based and semantic product features in context of domain concepts. Our recommendation is based on semantic compatibility between the products and user profile in context of use. We use probabilistic knowledge representation and reasoning framework to achieve robust and flexible results. The approach has been validated with fashion preferences of several individuals with a large collection of Sarees, an ethnic dress for women in Indian subcontinent. © 2013 IEEE.

Proceedings - 2013 IEEE/WIC/ACM International Joint Conference on Web Intelligence and Intelligent Agent Technology - Workshops, WI-IATW 2013

An ontology based personalized garment recommendation system

High quality depth map estimation is required for better visualization of 3D views as there is great impact of depth map quality on overall 3D image quality. If the depth is estimated from conventional ways using two or more images, some defects come into picture, mostly in regions without texture. We utilised Microsoft Kinect RGBD dataset to obtain input color images and depth maps which also includes some noise factors. We proposed a method to remove this noise and get quality depth images. First the color and depth images are aligned to each other using intensity based image registration. This method of image alignment is mostly used in medical field, but we applied this technique to correct kinect depth maps by which one can avoid cumbersome task of feature based point correspondence between images. There is no requirement of preprocessing or segmentation steps if we use intensity based image alignment method. Second, we proposed an algorithm to fill the unwanted gaps in kinect depth maps and upsampled it using corresponding high resolution color image. Finally we applied 9×9 median filtering on implementation results and get high quality and improved depth maps. © 2013 IEEE.

2013 4th National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics, NCVPRIPG 2013

High quality depth map estimation by kinect upsampling and hole filling using RGB features and mutual information

This paper is aimed at exploring the potential of using discriminatory primitives containing words for the task of detecting skilled forgeries. We consider handwritten Devanagri documents for this work. We have obtained experimental handwriting data from subjects who have contributed handwriting samples in their natural handwriting. Other authors are asked to imitate the writing style of the subjects to produce a skilled forgery sample. Most of the literature dealing with writer recognition focus on signatures and very few reports have addressed the problem of detecting forgeries for handwritten Indian scripts. We also use multiple words based classification for the targeted task of forgery detection. Our experiments show encouraging results. © 2013 IEEE.

Off-line skilled forgery detection on handwritten Devanagri script

Writer recognition based on peculiarity of hand-writing is an important aspect of any forensic analysis. We present an approach for selecting best discriminative primitives for writer recognition. After selecting the primitives we also propose a hybrid system by combining both writer recognition and handwriting recognition for improved accuracy. We have also validated the performance of selected primitives on publically available dataset. We have performed this study on the Devanagri script. Experimental results verified the effectiveness of the proposed franework. © 2013 IEEE.

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

Most discriminative primitive selection for identity determination using handwritten devanagari script

This paper presents a novel parameterized variety-based 3D exploration model to comprehend the sparse unstructured collection of photographs, and automatically plan virtual 3D tours of the world's landmarks through interesting viewpoints without explicit 3D reconstruction. The proposed system analyzes the collection of unstructured but related image data containing the same location or environment to create a parameterized scene graph: a data structure that conveys spatial relations and enable smooth virtual navigation between photos. A novel statistical-heuristic criteria is evolved exploiting the scene spatial layout and appearance to automatically identify best available portals between photographs. Once well connected, the graph is parameterized and consistently rendered choosing visually compelling 3D transition paths, maintaining a pleasing essence of parallax. The system's ability is demonstrated on several casually captured personal photo collections of heritage sites and imagery gathered from 'Flickr' data. © 2013 IEEE.

3DTV-Conference

A novel variety-based 3DTV content generation scheme for casually captured sparse photo collections

The ability to exert real time, adaptive control, of the transportation process is the core of an intelligent traffic system. We propose a video based adaptive traffic signaling scheme for reducing waiting period of vehicles at road junctions without detecting or tracking vehicles. The traffic signal timing parameters at a given intersection are adjusted automatically as functions of the local traffic conditions. The video sequences recorded at junctions are used for generating Spatial Interest Points (SIP) and Spatio-Temporal Interest Points (STIP). The traffic congestion at the junction is estimated using SIP and STIP. The decision rules are based on a definitive analogy between road traffic and computer data traffic wherein road vehicles are compared with data packets on the network. The system is similar in approach to the technique of Weighted Round Robin (WRR) queuing, a scheduling discipline used in data communication networks. Local traffic information is used to adjust the phase split keeping the cycle time constant. Two methods have been proposed. The first method, Optimal Weight Calculator (OWC), minimizes traffic at an intersection by determining the optimal phase splits or weights. The second method, Fair Weight Calculator (FWC), calculates weights relative to the road with minimum traffic to bring more fairness. After applying the respective algorithms mathematically on varying traffic conditions, OWC was found to be more equitable in the allocation of green time which is suitable for highly weight-sensitive junctions. For traffic with road priorities, FWC was found to be a much faster and effective control strategy. © 2013 IEEE.

2013 7th International Conference on Distributed Smart Cameras, ICDSC 2013

Video based adaptive road traffic signaling

This paper presents an architecture of a smart space built with robots and distributed smart cameras connected via a network. We present a framework for tracking and recording presence of objects over a large space using vision as the basic sensing mechanism. A MAP based object identification scheme which works on Grassmannian manifold has been presented. Experimental results establish validity of the approach. © 2013 IEEE.