KSVM-Based Fast Intra Mode Prediction in HEVC Using Statistical Features and Sparse Autoencoder

Abstract

High Efficiency Video Coding (HEVC) is designed to deliver a video communication with better quality at reduced bit rate. For intra coding, HEVC employs an effective hierarchical quad tree partitioning and an exhaustive optimal mode search which increases the time complexity. Aiming this issue, we propose a Support Vector Machine (SVM)-based method to effectively predict the intra mode. Compared to the standard HEVC encoder HM-15.0, the proposed method could reduce 57.6% of encoding time at a bit-rate penalty of 3.3% at an average PSNR decline of only around 0.09 dB.

Keywords

• HEVC
• intra mode prediction
• RMD
• SVM
• Hu moments
• sparse autoencoder