Machine Learning-based Fast Intra Coding Unit Depth Decision for High Efficiency Video Coding

This paper proposes a fast coding unit (CU) depth decision algorithm for intra coding of high efficiency video coding using an artificial neural network (ANN) and a support vector machine (SVM). Machine learning provides a systematic approach for developing a fast algorithm for early CU splitting or termination to reduce intra coding computational complexity. Appropriate features for training SVM models were extracted from spatial and pixel domains of the current CU. These features were classified into three types for three SVM training models at each depth, and different weights were assigned on the basis of the ANN analysis. Experimental results showed that the proposed fast algorithm saves at most 48.5% and on average 33% encoding time with a 1.55% Bjøntegaard delta bit rate (BDBR) loss compared with HM 15.0.