A Fast HEVC Inter CU Size Decision Algorithm Based on Multi-Class Learning

摘要

HEVC achieves much higher coding efficiency compared with previous coding standards at cost of significant computation complexity. This paper proposes an efficient machine-learning based inter CU size decision algorithm for HEVC. The proposed algorithm uses a wrapper method based on F-score to select features in several factors that may be related to the CU partition, and designs a three-output classifier to control the risk of false prediction by combining the machine learning classifier with the RDcost judgment of the original algorithm. Experimental results show that the proposed method provides a reduction of 45.64% on average in computational complexity with a small increase in the bit-rate and a negligible reduction in PSNR values.