Quadtree Degeneration for HEVC

Abstract

The quadtree is one of the most advanced techniques contributing to the excellent compression performance of high efficiency video coding (HEVC). However, the computational complexity increases because the quadtree examines all coding unit (CU) sizes to obtain the optimal CU partitioning. This paper focuses on quadtree degeneration based on a proposed quadtree probability mechanism. Two techniques, a quadtree probability model (QPM) procedure and a quadtree probability update (QPU) procedure, are proposed. The QPM process estimates a CU distribution model based on a quantization parameter (QP) and a group of pictures (GOP). Based on the model, a new quadtree is constructed by skipping low probability tree nodes. The QPU process is performed to update the new quadtree based on scene content change. Update addresses model distortion and ensures the accuracy of the new quadtree. Experimental results demonstrate that the proposed quadtree probability mechanism for quadtree degeneration considerably reduces average encoding time (27.55%) for the low delay condition. Applied to lossless coding, the proposed mechanism achieves a significant 43.10% encoding time reduction. The experiments also show that the proposed quadtree probability mechanism improves HEVC coding efficiency for a variety of applications and sequence characteristics.