摘要
立体图像具有2倍于普通图像的数据量,给存储和传输带来了一定的困难,因此对立体图像的压缩和编码非常重要.文中讨论了Delaunay三角形模型基立体图像编码的各个环节,提出一种可以保持网格拓扑结构不变的视差估计方法.该方法是一个三步迭代过程:首先利用迭代块匹配算法得到较精确的匹配结果;然后利用迭代多边形匹配算法对得到的匹配结果进行优化,使得一些三角形的预测误差最小;最后删除掉视差不可靠的节点.利用变长编码来编码网格节点和视差矢量,并对残差图像利用DCT进行编码.实验结果表明,与基于块的传统编码方法和其他网格匹配算法相比,该方法在获得较高压缩比的同时,解码图像亦具有较高的主客观质量.
Compression of stereo image is important since its data is in double size compared with the corresponding non-stereo image in terms of storage and transmission. Through an analysis to the processing stages of stereo image coding based on Delaunay triangulation model, a disparity estimation algorithm, capable of maintaining the topology structure of the mesh, is proposed. The algorithm is implemented by a three-stage iterative process. At the first stage, an iterative block matching algorithm is applied to generate more accurate initial disparity vectors, and at the second stage, an iterative polygon matching algorithm is employed to optimize the initial disparity vectors obtained at the first stage to minimize the error of certain triangulations. Finally at the third stage, the unreliable nodes are deleted. Variable length code is used to encode the mesh nodes and disparity vectors, and the residual image is coded with DCT. Experimental results show that the proposed algorithm can provide better subjective and objective quality compared with the conventional block coding scheme and other mesh-based matching methods under a comparable high compression ratio.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2008年第12期1575-1582,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60472083
60872141)
关键词
立体图像编码
视差估计
DELAUNAY三角形网格
六参数映射
残差图像
stereo image coding
disparity estimation
Delaunay triangulation mesh
six parameters affine transformation residual image
