High-Speed CTU-Based HEVC Intra-Coding via Deep Learning Technique

Abstract

In recent years, High-Efficiency Video Coding (HEVC) standards have gained widespread popularity owing to their ability to provide improved video compression performance. Nevertheless, the HEVC intra-coding procedure is still computationally demanding, which prevents its real-time application on devices with limited resources. In order to overcome this difficulty, we provide a deep learning-based method for quick CTU-based intra-coding in HEVC that makes use of a unique Convolutional Neural Network (CNN) to forecast intra-coding modes. Our method achieves an average encoding time reduction of 68.69% on the RAISE dataset and 62.26% on the JCT-VC test set, with a BD-BR increase of 1.238% and BD-PSNR loss of -0.084 dB, outperforming LeNet-5 and OSDN. This advancement enhances video coding efficiency for real-time applications.