VISUAL INFORMATION TECHNOLOGY
Stereo vision attempts to recover 3D structure of the scene from images. A fundamental problem in stereo is the computation of pixel correspondences. Several schemes for pixel correspondence, which integrate different types of information present in the image, have been devised. A novel strategy to improve the correspondence by adaptively weighing various supports to compute the correspondence has been developed. The supports can be multiple windows, multi-base line images or multiple features. The feature integration scheme is being extended to other image processing applications such as document image analysis, dynamic video processing, video-stereo, etc.
A flexible scheme for indexing and retrieval has been developed. Existing algorithms for indexing image databases are often tuned to a specific application. With the emergence of large and diverse image collections, one could easily come across new and challenging requirements. With new low cost storage devices, digital libraries are becoming more and more popular. Automatic indexing and retrieval from this collection is being investigated. A typical application will be to index TV news clips for better retrieval at latter point of time. Audio-visual clues are employed for segmentation. Soft computing tools will be used in the feature extraction phase.
Document image analysis research requires a closer look in the Indian context. Indian languages have different scripts and their mixed presence makes the problem of analysis and recognition of Indian language documents difficult. Work on segmentation and recognition of Indian languages in document pages and Internet images and online character recognition system for Indian languages has commenced. The applications can be for indexing document pages or web images, understanding documents or effective compression of the documents. Character recognition is another important problem in this domain.
Virtual Reality is an investment intensive area. A low-cost model for VR that uses inexpensive PCs connected to an expensive, central client for full VR experience, using Image Based Rendering techniques at the client's end is under development. Work on novel representation schemes suitable for large virtual environments is in progress. This scheme represents the world as a number of visibility-limited models, and not as a single global model. Applications of visualization research can be many. Scientific data visualization has applications in meteorology, oceanography, molecular chemistry, etc. Tools for effective visualization in collaboration with the scientific community using graphics and virtual reality are being developed.
Product development using advanced technologies from image processing, computer vision, graphics, etc, is also being done. These could be for video indexing, segmentation and retrieval, biometric systems, online handwriting recognition for Indian languages, etc.