DIMD Histogram Adjustment for Screen Content Coding

In this paper, we investigate on enhancing the coding performance of DIMD (Decoder-Side Intra Mode Derivation), one of the intra prediction tools under consideration for next generation video coding standard beyond VVC capability, for a screen content video which is often referred to as computer-generated video. We note that the combination of the intra-predictors of the planar mode and the directional modes derived based on the histogram of gradients (HoG) in current DIMD can result in a blurry predictor and it can be specially a problem for screen content video that typically has many sharp edges, and in this context we devise a novel way of generating a sharper intra-predictor by emphasizing dominant directional modes and their neighboring modes in DIMD. Compared to the existing DIMD in the test model, ECM 10.0 under the all intra configuration, we observe coding performance improvements of -0.02%, 0.05%, and -0.02% respectively in the Y, Cb, and Cr channels for the class F, and -0.01%, 0.01%, and -0.09% for the class TGM.