Decoder-Side Intra Mode Derivation Based on a Histogram of Gradients in Versatile Video Coding

Abstract

Following the finalization of the state-of-the-art video compression standard High Efficiency Video Coding (HEVC) in 2013, there is a growing need for more compression efficiency to handle various new video services and formats. In order to address this need, the ITU-T and MPEG standardization bodies are working together in a Joint Video Expert Team (JVET) group on the standardization of a new video codec called Versatile Video Coding (VVC), aiming to achieve 50% bitrate reduction over HEVC by 2020. One of the solutions that can be proposed in the intra mode is increasing the number of prediction modes, thus allowing more efficient predictions and less energetic residues. However, this solution causes an increase in the signaling cost of the chosen mode, which counterbalances some of the gains obtained by the newly added prediction modes. To address this problem, we propose to derive prediction modes at the decoder, to avoid sending them in the bitstream. This is done by deriving the best prediction mode at the encoder through a gradient calculation performed on each pixel of a well-determined causal region of the image. This method is identically applied at the decoder, thus avoiding the intra mode signaling in the bitstream. This process defines a new coding mode named Decoder side Intra Mode Derivation (DIMD), which competes with other coding modes normally tested at the encoder, including the classical intra coding mode. The proposed method reduces, on average 0.3%, the bitrate used to encode a sequence configured in All Intra (AI), with limited additional decoding complexity (2%).