基於深度學習之改良式多功能影像編碼快速畫面內模式決策研究

Improved Fast Intra Mode Decision in H.266/Versatile Video Coding (VVC) Based on Deep Learning

H.266/Versatile Video Coding (VVC) 是針對 4K 以上的超高畫質影片，且能適用在高動態範圍(High Dynamic Range Imaging, HDR)及廣色域(wide color gamut, WCG)中，但基於四元樹加二元樹(Quadtree plus Binary Tree, QTBT)的編碼單元(Coding Unit, CU)結構增加了 H.266/VVC 編碼的計算複雜性。本論文提出了一種基於深度學習之改良式多功能影像編碼快速畫面內模式決策方法，減少 H.266/VVC 內編碼複雜性以加快H.266/VVC 的編碼速度，並將畫面內影像編碼結合卷積神經網路(Convolutional Neural Networks, CNN)在 H.266/VVC 畫面內編碼的模式預測決策，以達到比原始編碼方式(JEM7.0)更好的編碼效能。

 

H.266/VVC is ultra-high-definition video over 4K, and can be applied in High Dynamic Range Imaging (HDR) and wide color gamut (WCG). However, it has high coding computational complexity based on the coding unit (CU) structure of a quadtree plus binary tree (QTBT). This plan first proposes a fast coding unit spatial features decision method to reduce the coding complexity in H.266/VVC such that the H.266/VVC coding can be speed up. Another important contribution of this plan is to combine video coding with Convolutional Neural Networks (CNNs) in H.266/VVC in-frame coding mode prediction decision. It can be shown that the proposed methods can achieve better encoding performance than the original encoding method (JEM7.0).