Region-of-interest based H.264 encoder for videophone with a hardware macroblock level face detector
Region-of-interest based H.264 encoder for videophone with a hardware macroblock level face detector
复制标题
DOI：
10.1109/mmsp.2009.5293338
复制
发表时间：
2009-10
期刊：
影响因子：
--
通讯作者：
T. Zhang;Chen Liu;Minghui Wang;S. Goto
中科院分区：
文献类型：
--
作者：
T. Zhang;Chen Liu;Minghui Wang;S. Goto
感兴趣区域（ROI）可以应用于H.264视频编码器，提高主观质量，降低计算复杂度。针对低成本硬件实时编码器在以人脸为ROI的可视电话中的应用，本文提出了一种人脸检测算法，将每个宏块（MB）检测为人脸的一部分或非人脸的一部分。该人脸检测算法具有独特的估计和验证过程，并且可以通过MB级流水线架构与H.264编码器结合使用。人脸中97.91%的mb被检测到。设计了人脸检测算法的VLSI体系结构，实现了4.3k的栅极面积。在100MHz时，功耗仅为1.45mW。提出了一种基于ROI的动态参数H.264编码器，以提高主观质量，降低RDO复杂度。在相同的比特率下，ROI中的PSNR提高了4.8dB。在类似视频电话的序列中，编码时间减少到54.4%。
Region-of-interest (ROI) can be applied in H.264 video encoder to enhance subjective quality and reduce computation complexity. For the aiming application of low cost hardware real-time encoder in videophone with faces as ROI, this paper proposes a face detection algorithm to detect each macroblock (MB) as one part of a face or not. This face detection algorithm has a unique estimation-and-verification process and can be combined with a H.264 encoder by MB level pipeline architecture. 97.91% MBs in faces can be detected. VLSI architecture of proposed face detection algorithm is designed and an area of 4.3k gates is achieved. Power consumption is only 1.45mW at 100MHz. A ROI based H.264 encoder with dynamic parameters is proposed to enhance subjective quality and reduce the rate-distortion-optimization (RDO) complexity. The PSNR in ROI increases for 4.8dB under similar bit rate. Encoding time is reduced to 54.4% in videophone-like sequences.