 |
Research and Development (2005-2010)
Image Processing in the block DCT domain
Various algorithms on image resizing, color enhancement, color constancy, and image filtering have been developed in the block DCT domain. A simple but effective algorithm by scaling DCT coefficient leads to the color enhancement of images (color enhancement by scaling (CES)), while color constancy operations are mostly carried out with the DC coefficients of the transformations. The latter is found to be quite efficient as it handles less data compared to that in spatial domain operations, yet providing comparable results. The techniques for image filtering are developed exploiting the convolution-multiplication properties of mixed-DCTs and DSTs. Algorithms for equivalent computations of linear convolution in the block DCT domains are also developed (reported only in the book "Image and video processing in the compressed domain (CRC Press, 2011)") for separable / non-separable, and symmetric / anti-symmetric / asymmetric filters.
Image and Video Transcoding
Using image filtering in the block DCT domain transcoding algorithms of images from DCT to DWT and vice-versa have been developed. For videos, new techniques for inverse motion compensation have been proposed and used for resizing MPEG videos. Similarly, by using motion vectors of a compressed stream, a transcoding technique from H.264 to MPEG-2 encoding has been proposed.
Video compression
A new video compression algorithm exploiting the higher compressibility of JPEG2000 and motion compensated prediction, and preciseness in rate control of JPEG2000, has been developed. The technique is called Motion Compensated JPEG2000 (MCJ2K) video compression and is found to achieve much higher compression of existing Motion-JPEG2000 scheme at high rate encoding zone, yet providing comparable quality of video. There were also efforts in enhancing the speed of video compression algorithms, in particular, in the computation of motion vectors. The techniques by exploiting the reduction of computation of sum of absolute differences (SAD) using less number of representative points in a macro-block, guiding search to the neighborhood of most-likely minimum error blocks and predicting skip blocks during encoding, are developed in this regard.
Video modeling and processing
Algorithms for analyzing broadcast soccer video and echocardiograms have been developed. The analysis follows the same state based video modeling as proposed previously. A dynamic programming based object tracking algorithm for detecting the trajectory of ball, players, etc. has been developed and used for summarizing soccer videos. In analyzing echocardiogram, synthetic generation of M-mode echocardiograms is found to be useful. For this purpose, two types of scanning, namely translational sweep and radial scan, have been used.
Image Steganography
Different new algorithms for image steganography have been developed in spatial and DCT / DWT domains. These algorithms are found to be providing better performance than some of the existing algorithms against different steganalytic attacks. The approaches include using varying bases of number systems, pixel swapping, randomized cropping, etc.
Visual Saliency Model
A graph based visual saliency model (ViSaNet) has been proposed and ground truth data for testing different saliency models are collected. The model has been used for image retargeting in hand held devices.
Digital Distance
A new class of distance functions, called weighted t-cost distance functions, has been proposed. This class of distances generalizes previously known classes such as m-Neighbor distance functions, octagonal distance functions and t-cost distance functions. A typical member of this class (called inverse square root weighted t-cost distance) is found to be a good approximator of Euclidean metric in any arbitrary dimension.
Medical Instrumentation
A data acquisition system for capturing different types of bio-medical signal such as ECG, PPG, and Cardio-phonogram, has been developed. The system is used specifically for carrying out clinical study of PPG signals in different groups of patients, such as hypertensive, diabetic, and both. A new model for characterizing PPG signals has been also proposed and used in this study. A design and implementation of cordless ECG acquisition system has also been carried out as a proof of concept.
Neo-natal patient care
A prototype system for management of critical care of new-born babies in a Neonatal Intensive Care Unit has been developed and deployed in the Department of Neonatology of the SSKM Hospital, Kolkata. For follow-up patients, image and video processing based techniques are being developed to aid the Hammer-Smith Infant Chart based analysis. A technique for adductor measurement has been developed in this regard. There are also efforts for analyzing EEG signals and USG brain images of new-born babies.
Telemedicine Systems
A telemedicine system for HIV-pediatric patients has been developed and deployed in the Calcutta Medical College, Kolkata. The system has been integrated in a web-based telemedicine system, called iMediK. iMedik has all the features of TelemediK, a peer to peer system developed earlier. It is easy to install and works also in a public network. It has a four layered architecture providing better data and network security compared to other existing models. The design is also extended to different distributed models of public health delivery. iMediK is also installed in the GB Pant Hospital and Medical College, Agartala, Tripura.