• Document Image Understanding
• Content Based Retrieval of Multimedia Data
• Geometry of Collections in Computer Vision
• Graphics, Visualization and Virtual Reality
• Retinal Image Processing
• Pattern Recognition and Machine Learning
• Dynamic Scene Analysis
• Biometrics
• Other Topics

CVIT is building tools for document understating tasks with special emphasis on Indian languages. These include extensible multi-lingual optical character reader (OCR) systems, form processing systems, document database systems for scanned documents, reading aid for printed text, and on-line character recognition systems. Advanced prototypes of some of these systems have been demonstrated at the Institute's open house and other forums. The centre believes that the rich, multi-lingual scenario that exists in the country provides special challenges to document understanding research.

Specific activities in this area include:

• OCR System: Devanagari, Telugu, Malayalam, Bengali
• Offline and Online Handwriting Recognition
• Sanjaya Reading System for the Blind
• Form Processing System
• Document Database System

Content-based retrieval of relevant items from a large collection of images and video is a challenging problem. We are building a system that addresses some aspects of this problem. Issues related to the adaptation, advanced and natural querying schemes, feature selection, content representation, etc., are being explored at the centre. Domain specific systems for News Videos, Human faces, Medical Images, Logo Images, etc., are quite possible today and we are building them.

Specific activities in this area include:

• Rule-Based Image Retrieval System
• Domain Specific Image Retrieval System
• Learning in Image Retrieval
• Interactive Large Scale Retrieval from Real Databases
• Privacy in Image Retrieval

Understanding of dynamic and static objects from its multiple images is an important issue in computer vision systems. Geometric constraints on multiple views of the same scene has been an active area of research in Computer Vision recently. We have developed a new framework for recognition of some classes of objects and demonstrated its applications on real-life images. We are currently exploring the problem of tracking moving objects in multiple views and compression of the multiview video data for virtual presence.

Specific activities in this area include:

• Shape Analysis in Multiple Views
• Moving Points in Multiple Views

Computer Graphics and Virtual Reality are areas with wide applications in training, engineering, etc. The centre is active in research into the visualization and navigation of large virtual environments, representation and visualization of molecular data, developing client-server models for inexpensive Virtual Reality, image-based rendering, and the study of software architectures for next-generation VR systems.

Specific activities in this area include:

• Terrain Processing
• Programming GPUs for Graphics and GPGPU
• VR-based laparoscopy surgery training tool
• Graphics subsystem for molecular visualization of BioSuite ((jointly with TCS)
• Client-server geometry streaming model for remote visualization
• Depth Movie Compression and Transmission
• Scalable Display Walls

The centre works on processing of medical images, especially images of retina of the eye.

Specific activities in this area include:

• Retinal Image based Disease Detection

The areas of Pattern Recognition and Machine Learning encompasses a wide variety of topics that are essential for most visual information processing problems. Our primary areas of interest include classifier design and systems that learn from large collections of data.

Specific activities in this area include:

• Design of highly accurate and efficient classifiers
• Learning from Large Document Collections
• Learning from User Feedback

The area of Dynamic Scene Analysis includes the processing, classification, understanding and summarisation of video streams from both single and multiple cameras.

Specific activities in this area include:

• Activity Recognition
• Human Detection and Tracking

Biometrics deals with the identification of people based on their physiological or behavioural characteristics, such as fingerprints, face, hand geometry, handwriting, speech, etc. The primary focus of research are on improving the performance of popular biometric modalities and to develop computer vision based solutions to biometric problems.

Specific activities in this area include:

• Camera-based Biometrics
• Enhancing Weak Biometrics
• Writer Identification

In addition to the above, our centre works on a variety of problems in various topics. Some of them are listed below:

• Mosaicing of Arial Images
• Automated Inspection Systems