Adaptive segmentation-based filtering of reference samples for intra prediction

In this paper, we present a new adaptive filter of reference samples for increasing the RD-efficiency of the intra-prediction modes adopted for the HEVC/H.265 standard. This filter is based on a combination of different techniques and aimed at reducing (or even removing) artifacts of different types such as blockiness and impulse noise generated in an image sensor. In addition, the strength of intra-smoothing is adjusted in accordance with the smoothness of the blocks that the reference samples belong to. The proposed filter provides the coding gain of 1.8% for luma and 2.4% for chroma, on average.