Template-based intraprediction using multiple reference lines

A novel template-based multiple reference line intra prediction method (TMRL) is presented in this paper. This new method comprises three steps. First, a candidate list is constructed, where each entry consists of an intra prediction mode and a non-adjacent reference line index. Second, the entries in the candidate list are reordered according to a cost function defined and calculated over a specified template area. The process of constructing and reordering of the candidate list is done both the encoder and decoder sides. As the third step, the encoder selects a candidate producing the best intra prediction and encode the corresponding candidate index into the bitstream. For the decoder, upon completing the first two steps, it uses the candidate index decoded from the bitstream to fetch the corresponding candidate and generates the prediction block according to the intra prediction mode and reference line given by the fetched candidate. TMRL has been evaluated by Joint Video Expert Team (JVET). The simulations conducted under the JVET common test conditions have demonstrated impressive coding gains with a minor encoder and decoder complexity increase. JVET has adopted TMRL into its reference software as a candidate tool for the purpose of future video coding standards.