Examination Committee

Prof Alvin Yiu-cheong SO, SOSC/HKUST (Chairperson)
Prof Weichuan YU, ECE/HKUST (Thesis Supervisor)
Prof Gene CHEUNG, ECE/HKUST (Thesis Co-supervisor)
Prof Moncef GABBOUJ, Department of Signal Processing, Tampere University of Technology (External Examiner)
Prof Zhiyong FAN, ECE/HKUST
Prof Lu FANG, ECE/HKUST
Prof Gary S H CHAN, CSE/HKUST

Abstract

Efficient video/image coding techniques are desired for the dramatically growing demand of storage and transmission of visual information. The widely-used coding standards, e.g., H.264 and HEVC, are all based on a hybrid video coding framework which consists of a predicting process followed by transform coding and entropy coding. Considering the results of encoding one image, the discontinuous regions, i.e., regions containing edges/contours, consume most of the coding bits because of the inaccurate prediction of the discontinuities. Instead of using an integrated predicting technique to encode both the discontinuous regions and the remaining regions, an alternative way is to perform a separate process to the discontinuities in the image. Separate discontinuity handling provides more freedom to design an algorithm for efficient encoding of the discontinuous regions.

In this thesis, we focus on handling the discontinuities of the video/image in two directions to improve the overall coding efficiency. We first discuss a contour assisted coding framework which pre-encodes the discontinuities before feeding the input image into a conventional encoder, and propose efficient algorithms to encode the discontinuities. Then we propose a new approach, adaptive coding order, for a hybrid video coding framework to reduce the discontinuities.

For a contour assisted coding framework, we focus on the problem of lossless and lossy compression of detected contours (discontinuities) in the image. Specifically, to encode the symbol sequence contour, we compute an optimal variable-length context tree (VCT) via a maximum a posterior (MAP) formulation to estimate the symbols' conditional probabilities. For the lossy case, we design fast dynamic programming (DP) algorithms that optimally trade off the coding rate of an approximated contour given a VCT with two notions of distortion. Experimental results show that the proposed contour coding algorithms significantly improve the overall coding performance for different applications. To deal with the problem of compressing noised contours, we further propose a joint contour denoising/compression algorithm based on a burst error model.

For the direction of adaptive coding order, the order of processing the basic coding unit, a block of pixels, in hybrid video coding is adaptively adjusted based on the edge direction rather than pre-determined in a fixed manner. Specifically, determining the block coding order is formulated as a travelling salesman problem that is solved using the DP algorithm. Experimental results show that the proposed algorithm outperforms the state-of-the-art HEVC.