Adaptive cross-component prediction for 4:4:4 high efficiency video coding

The work on support for higher bit depths and 4:2:2 as well as 4:4:4 chroma sampling formats for the High Efficiency Video Coding (HEVC) standard is currently being conducted under the term Range Extensions (RExt). A technique that exploits the correlation between residual color components in 4:4:4 chroma sampling format, also referred to as Cross-Component Prediction (CCP), has been adopted as part of the current RExt draft. In this paper, this relatively simple but yet effective CCP scheme is presented. Conceptually, CCP relies on the idea that an adaptively switched predictor based on a linear model is invoked for coding of the residuals of the second and third color component by using the residual of the first color component. Experimental results show that, depending on the underlying color space, average bit-rate savings in the range of 2-18% or 3-26% can be achieved by CCP for test sets of natural and screen content, respectively.