Date of Award

Spring 1997

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering - (Ph.D.)

Department

Electrical and Computer Engineering

First Advisor

Yun Q. Shi

Second Advisor

Joseph Frank

Third Advisor

Edwin Hou

Fourth Advisor

Daochuan Hung

Fifth Advisor

Zoran Siveski

Abstract

Recently, many activities of the International Telecommunication Union (ITU) and the International Standard Organization (ISO) are leading to define new standards for very low bit-rate video coding, such as H.263 and MPEG-4 after successful applications of the international standards H.261 and MPEG-1/2 for video coding above 64kbps. However, at very low bit-rate the classic block matching based DCT video coding scheme suffers seriously from blocking artifacts which degrade the quality of reconstructed video frames considerably. To solve this problem, a new technique in which motion compensation is based on dense motion field is presented in this dissertation.

Four efficient new video coding algorithms based on this new technique for very low bit-rate are proposed. (1) After studying model-based video coding algorithms, we propose an optical flow based video coding algorithm with thresh-olding techniques. A statistic model is established for distribution of intensity difference between two successive frames, and four thresholds are used to control the bit-rate and the quality of reconstructed frames. It outperforms the typical model-based techniques in terms of complexity and quality of reconstructed frames. (2) An efficient algorithm using DCT coded optical flow. It is found that dense motion fields can be modeled as the first order auto-regressive model, and efficiently compressed with DCT technique, hence achieving very low bit-rate and higher visual quality than the H.263/TMN5. (3) A region-based discrete wavelet transform video coding algorithm. This algorithm implements dense motion field and regions are segmented according to their content significance. The DWT is applied to residual images region by region, and bits are adaptively allocated to regions. It improves the visual quality and PSNR of significant regions while maintaining low bit-rate. (4) A segmentation-based video coding algorithm for stereo sequence. A correlation-feedback algorithm with Kalman filter is utilized to improve the accuracy of optical flow fields. Three criteria, which are associated with 3-D information, 2-D connectivity and motion vector fields, respectively, are defined for object segmentation. A chain code is utilized to code the shapes of the segmented objects. it can achieve very high compression ratio up to several thousands.

Share

COinS