Novel Chroma Sampling Methods for CFA Video Compression in AVC, HEVC and VVC

Abstract

The images obtained by the camera sensors are stored in the format of CFA, Color Filter Array, which only contains limited information about the R, G, and B signals of images. The images in CFA are upsampled into RGB 444 format by demosaicking algorithms, converted into YCbCr 444, then chroma-downsampled into YCrCb 420 format for image/video compression. In this paper, a novel chroma-downsampling method is developed to improve a state-of-the-art method. The optimization problem for the proposed method considers more factors and provides a better closed-form solution for the optimally sampled chroma values. Reversely in the decoder, a chroma-upsampling method is required for the image reconstruction. In the proposed chroma-upsampling method, the neighboring chroma pixels and their distances to the recovered pixel are considered; the procedure is to produce weightings to redistribute the total pixel values in the original estimates. Combining the proposed chroma-downsampling and the proposed chroma-upsampling works, the proposed method outperforms the state-of-the-art method by BDPSNR (Bjontegaard Delta peak signal-to-noise ratio) 0.3161 for H.264/AVC, 0.3845 for HEVC (High Efficiency Video Compression) and 0.3739 for VVC (Versatile Video Coding), averaged over all tested videos and all 7 different considered CFA formats.