基于分层视差估计的立体图象编码

Stereo Image Coding Based on Hierarchical Disparity Estimation

基于立体视频数据压缩的目的 ,提出了一种基于分层视差估计 /补偿的立体图象编码方案 ,该方案是采用JPEG标准独立编码参考图象 ,并利用视差估计 /补偿技术编码目标图象 .应用分层马尔可夫随机场 (MRF)模型、率失真 (RD)模型以及交叠块匹配的混合视差估计 /补偿算法 ,可得到光滑准确的视差场 .与通常的变尺寸块匹配(VSBM)相比 ,该算法得到的视差补偿图象的峰值信噪比 (PSNR)可提高 2 .5 d B左右 ;双向半像素精度的视差估计/补偿的性能要比单向整像素搜索提高 3d B.实验结果表明 ,该立体图象编码方案能有效地压缩立体图象数据 ,并可推广到立体序列图象的编码压缩中 .

To compress the stereo video data, we propose an algorithm for stereo image coding based on hierarchical disparity estimation/compensation. A stereo image pair consists of a reference image and a target image. The reference image is coded by method using JPEG standard, while the target image is coded using the disparity estimation/compensation technique. We apply the wavelet transformation to code the created residual image, and use DPCM(difference pulse coding modulation) to code the disparity fields. Hierarchical Markov Random Field(MRF) model, Rate-Distortion(RD) model and the overlapped block matching/compensation are applied in the hybrid disparity estimation method to obtain a smooth and consistent disparity field. In terms of peak signal to noise ratio(PSNR) of the disparity compensated image, the proposed disparity estimation algorithm achieves about 2 5 dB higher PSNR, as compared to conventional variable size block matching. The experimental results also show that the disparity estimation/compensation in half pixel accuracy and dual-direction can obtain a 3dB higher performance than that of integer pixel accuracy only in horizontal direction. In the same quality of reconstructed image, our coding system can obtain 0 387 bits reduction in terms of coding bit rate of the target image compared to VSBM based scheme, and 1 003 bits reduction compared to the one using JPEG standard to code the target image. The proposed hierarchical model for disparity estimation effectively improves the efficiency of stereo image coding.