Parameter optimization in H.265 rate-distortion by single frame semantic scene analysis
Ahmed M. Hamza, Abdelrahman Emad Abdelazim, Djamel Ait-Boudaoud · Electronic Imaging · 2019
The H.265/HEVC (High Efficiency Video Coding) codec and its 3D extensions have crucial rate-distortion mechanisms that help determine coding efficiency. We have introduced in this work a new system of Lagrangian parameterization in RDO cost functions, based on semantic cues in an image, starting with the current HEVC formulation of the Lagrangian hyper-parameter heuristics. Two semantic scenery flag algorithms are presented and tested within the Lagrangian formulation as weighted factors. The investigation of whether the semantic gap between the coder and the image content is holding back the block-coding mechanisms as a whole from achieving greater efficiency has yielded a positive answer.