【摘要】 传统的立体图像编码方法多采用基于块的视差补偿算法 ,但这种算法不能更为准确地估计视差矢量 ,在低码率时重建图像的块效应比较严重 提出了一种新的基于Delaunay三角形网格的立体图像编码算法 ,该算法首先对参考图像进行自适应的三角形网格剖分 ,然后对三角形顶点在目标图像中进行视差矢量估计 利用三角形模型和 6参数仿射变换计算三角形内部各点的视差矢量 ,并产生残差图像 结合实验统计对文中算法产生的残差图像进行了分析 ,采用嵌入式DCT算法编码残差图像 从实验结果看 ,文中算法视差矢量估计的准确性较传统算法提高了 10 %左右 在相近比特率下 ,平均PSNR大约提高 1dB更多还原

【Abstract】 Traditional way of stereo image coding uses the framework of block based disparity estimation and disparity compensated residual coding But the estimated disparity vectors are inaccurate, and it tends to produce artifacts at block boundaries in the predicted image, especially when low rate coding is used The proposed DT mesh based stereo image coder uses adaptive DT mesh to extract “feature points”, which are used for disparity estimation Then, disparity vectors inside a triangle are computed using six parameter affine transform The residual image is encoded by EZDCT Simulation results show that the proposed scheme raised the accuracy of disparity estimation at least 10% as compared to conventional algorithms It improves average PSNR about 1?dB更多还原