一种模型基人脸视频编码参数压缩算法

Compression Algorithm for Model-based Facial Video Coding Parameters

面向极低码率下的人脸视频编码领域,提出一种模型基人脸视频编码参数压缩算法.首先采用基于几何归一化人脸规范化过程的卡洛南-洛伊变换对人脸纹理进行压缩,与采用基于经典规范化过程的卡洛南-洛伊变换的压缩算法相比,提高了信噪比约0.5db;接着根据人脸运动单元来构造人脸运动基向量,并据此来消除同一帧内人脸表情运动参数的相关性,进而结合帧间编码和帧内编码在无编码延迟下完成对人脸表情运动参数的压缩.实验证明了该算法能够显著地提升模型基人脸视频编码参数的传输效率.

For facial video coding at ultra-lowbit-rate,a compression algorithm for model-based facial video coding parameters was proposed. Firstly,with the Karhuner-Loeve（ KL） transform based on geometrical normalization,facial texture was compressed to obtain enhanced PSNR with around 0. 5db compared to that with KL transform based on classical normalization. Secondly,the correlation of facial animation parameter（ FAP） in the same frame was reduced by facial action basis function which was constructed based on facial action unit,thus FAP was compressed combining intra-frame coding and inter-frame coding without introducing any coding delay. Experiment results confirmed the algorithm can increase the transmission efficiency of model-based facial video coding parameters.