Improved quantization of DCT based compression algorithms for video teleconferencing applications.

Abstract

Compression schemes for the video sequences using the conventional non-content based quantization schmes seem to have reached saturation level in terms of the amount of compression. For this reason, some of the contemporary research in this field has shifted its focus to test the content-based adaptive quantization schemes to improve the perspective visual quality of the generated sequences at the given low bit-rates. In this thesis, two new schemes are studied for adapting the quantization process of the international standard H.261 algorithm at two levels. The first scheme adapts the quantization process at a macroblock level by adjusting the quantization-scaling factor (MQUANT) according to the importance of the regions of each image in the video sequence. The second scheme adapts the quantization process at a block level by modifying the quantization-weight matrix (wij) according to the direction of concentration of high energy DCT coefficients. These schemes deal specifically with the video sequences used for desk top face-to-face visual communications at low bit-rates of 128 Kbps. 64 Kbps and 28 Kbps. Improvements in the perspective visual quality are achieved for the two schemes.