Adaptive color-based coding tools for HEVC range extension

Abstract

High Efficiency Video Coding (HEVC) is the next generation video coding standard, which has been established jointly by the ITU-T Video Coding Experts Group and the ISO/IEC Moving Pictures Experts Group (MPEG) standardization organizations, known as the Joint Collaborative Team on Video Coding (JCT-VC). It is designed to reduce the bit-rate by half, with equal perceptual video quality compared to existing video coding standards. With the first draft of HEVC finalized in early 2013, several extensions, including range extensions, scalable extensions and 3D video extensions, have been discussed to fulfill various requirements of a broader range of applications. In this thesis, we firstly proposed to use a CU (coding block) level spatially adaptive KLT-based color transform, which achieves optimal energy compaction. The proposed algorithm transforms the current color space into a new color space, whose color channels are less correlated so as to improve the coding performance. The transform matrix is estimated from the reconstructed pixels from neighboring blocks. The adaptive transform tends to give color components with variable bit depth. As minimal change to the existing HEVC is desirable, all color components need to have 8-bit bit depth. Thus we also propose a novel dynamic range adjustment algorithm so that the existing HEVC Range Extension can be applied almost directly. Secondly, while rate-distortion (RD) optimization is well known to give optimal results, we propose two modifications to the RD cost function especially for the proposed adaptive KLT-based color transform so as to achieve superior performance. The first modification is to perform color adaptive scaling, which translates to using three color adaptive Lagrange multipliers. The second modification is to include in the cost function a term that considers the cross-color correlation among the distortions in different colors. Thirdly, we propose a palette-based compound image compression algorithm for artificial screen contents, such as those in video games and computer remote desktops. Screen content usually contains sharp edges, which often can not be coded efficiently with traditional transform coding. Palette-based methods use a palette plus index map technique to represent such blocks. To reduce the overhead in coding the locally adaptive color palette, the proposed algorithm uses a global color palette template to predict the local palette. Finally, in this thesis, the color characteristic among different bit streams, which are encoded from same video source but different quality levels, is investigated and a simplified generalized residual prediction (GRP) algorithm is proposed in scalable extensions of HEVC. The prediction of the current block in the enhancement layer is jointly predicted from the enhancement layer and also the base layer. Also, by studying the properties of merge mode and residual characteristics, the proposed algorithms simplify the coding process by only testing the GRP in merge mode. Experimental results show that the proposed algorithms improve the coding performance of the extensions in HEVC effectively. Besides color characteristic coding in HEVC, an image deblocking algorithm using convex optimization techniques is also presented.

Date of Award	2014
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology