Fast Mode Decision In Hevc Intra Prediction, Using Region Wise CNN Feature Classification
Shiba P. Kuanar, K.R. Rao, Christopher Conly · 2018
The computational time of HEVC encoder is increased mainly because of hierarchical quadtree based structure, recursive search for finding best coding units (CU), and the exhaustive prediction mode search upto 35. These advances improve coding efficiency but result in a high computational complexity. Therefore we propose \quad a convolutional neural network (CNN) based algorithm which learns the regionwise image features and performs a classification job. These classification results are later used in the encoder downstream systems for finding the optimal CUs in each of the tree blocks, and subsequently reduces number of prediction modes. For ourmodel training, we gathered a new dataset which includes diverse images for better generalization of our results. The experimental results show that our proposed algorithm reduces encoder time saving up to 66.89 % with a minimal BD-BR loss of 1.31% over the state-of-the-art machine learning approaches and reduces mode selection by 45.83 % with respect to HEVC baseline.