MULTI-STAGE PREDICTION SCHEME FOR SCREEN CONTENT BASED ON HEVC
Mundgemane Nagaraj, Nagashree
MetadataShow full item record
Screen content is a computer generated content which contains text, graphics and animations and the coding of such content is different from coding a photographic content. This thesis emphasizes on coding the screen content keeping the present video coding standard High Efficiency Video Coding (HEVC) as the base. An extension to HEVC to support screen content coding (SCC) is currently under development by the Joint Collaborative Team on Video Coding (JCT- VC) and the main goal of the extension is to provide improved compression performance of text, graphics and animation in addition to photographic content. This thesis focuses on a particular coding technique called palette mode to efficiently improve the coding abilities. The palette based coding takes the advantage of screen content having distinct limited number of colors and based on the colors and the structure of the content, the content is divided into palette table (color table) and palette index map. The palette index map represents the structure of the content and aims to exploit the correlation among the structural components. After coding the neighboring indices, it is observed that there are repeated patterns that are present in the structure and they may provide potential coding efficiency if repeated patterns are identified and coded using already coded patterns. This gives rise to two-stage prediction and coding schemes for screen content focusing to achieve bit-rate savings. If the coding of neighboring indices takes the first stage of coding, the second stage is of coding the non-local correlation of the repeated patterns. The proposed algorithm for the second stage coding combines the intra block copy full frame search mode for identifying the repeated pattern and palette mode to code the matched pattern. The proposed algorithm was implemented on standard test sequences and it provides average bitrate savings of 2.5% for all intra and 1.86% for random access profiles for lossless coding.