Fast CU Decision Algorithm Based on CNN and Decision Trees for VVC

Compared with the previous generation of High Efficiency Video Coding (HEVC), Versatile (VVC) introduces a quadtree and multi-type tree (QTMT) partition structure nested multi-class trees so that coding unit (CU) can better match video texture features. This makes compression efficiency VVC significantly improved, but computational complexity is also increased, resulting in an increase encoding time. Therefore, we propose fast CU decision algorithm based on DenseNet network (DT) classifier to reduce save more We extract spatial feature vectors model. Spatial are constructed by predicting boundary probabilities 4 × blocks 64 units. Then, using features as input DT classifier, through classification function model, top N division modes higher prediction probability selected, other skipped complexity. Finally, optimal mode selected comparing RD cost. Our proposed achieves 47.6% time savings VTM10.0, while BDBR only increases 0.91%.