Many of the current video compression algorithms perform analysis and coding operations in a block-wise manner. Most of them use a motion compensated DCT algorithm as the basis. Many other codecs, mostly academic and in their infancy and known as Second Generation techniques, utilize region and contour based and model based techniques. Unfortunately, these second-generation methods have not been successful in gaining widespread acceptance in both the standards and the consumer world. Many of them require specialized computationally intensive software and/or hardware. Due to these shortcomings, current block based methods have been finetuned to get better performance at even very low bit rates (sub 64 kbps). Block based motion estimation is the principal mechanism used to compensate for motion between frames in an image sequence. Although current algorithms are fast and quite effective, they fail in compensating for uncovered background areas in a frame. Solutions such as hierarchical motion estimation schemes do not work very well since there is no reference in past, and in some cases, future frames for an uncovered background resulting in the block being transmitted as an intra frame (which requires the most bandwidth among all type of blocks). This thesis intro duces an intermediate stage, which compensates for these isolated uncovered areas. The intermediate stage uses a localized decorrelation technique to reduce frame to frame temporal redundancies. The algorithm can be easily incorporated into exist ing systems to achieve an even better performance and can be easily extended as a scalable video coding architecture. Experimental results show that the algorithm, used in conjunction with motion estimation, is quite effective in reducing temporal redundancies.
Library of Congress Subject Headings
Video compression; Computer algorithms
Department, Program, or Center
Electrical Engineering (KGCOE)
Nadarajah, Dinesh, "Localized temporal decorrelation for video compression" (1998). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus