Fast CU partitioning algorithm for HEVC intra coding using Data Mining

Autores UPV
Año
Revista Multimedia Tools and Applications An International Journal

Abstract

HEVC adopts flexible coding unit (CU) partitioning by applying recursive CU splitting into four sub-CUs, up to four depth levels, which causes a significant complexity increase. Intra-prediction coding in HEVC achieves high coding performance through the exhaustive evaluation of all available CU sizes, with up to 35 prediction modes for each CU, selecting the one with the lower rate distortion cost. This work presents a novel CU size classifier comprising an offline-trained decision tree with three hierarchical nodes. The decision rules computed in each node are based on the content texture properties of CUs as well as the inter-sub-CUs statistics of the same depth level. Our approach can reduce the number of CU sizes to be checked by the Rough Mode Decision and Rate Distortion Optimization stages of intra-prediction coding. The experimental results show that the proposed algorithm can achieve over 50% coding time reduction, with no quality penalty in terms of the Peak Signal to Noise Ratio, and just a low bit rate increase (2%) compared with the HEVC reference model. A performance comparison with state-of-the-art proposals shows that this algorithm surpasses the best proposal in terms of time reduction, for the same coding performance penalty.