摘要
传统基于像素的立体匹配算法误匹配率较高。为解决该问题,提出一种基于图像区域分割和置信传播的匹配算法。采用均值偏移对参考图像进行区域分割,通过自适应权值匹配计算初始视差图,对各分割区域的初始视差用平面模型拟合得到视差平面参数,使用基于区域的改进置信传播算法求得各区域的最优视差平面,从而得到最终视差图。与全局优化的经典置信传播算法和图割算法的对比实验结果表明,该算法能降低低纹理区域和遮挡区域的误匹配率。
As the high error matching rates of the traditional pixel-based matching algorithm, a stereo matching algorithm based on image region segmentation and Belief Propagation(BP) is proposed. The mean shift algorithm is applied to segment the reference image into regions with homogeneous color, and the initial disparity of each pixel is calculated by means of the adaptive weights approaches. The disparity plane parameters are collected by plane model fitting on each segmented region. The ultimate disparity map is acquired by calculated the regional optimal disparity plane, which uses the improved region-based belief propagation algorithm. Compared with the pixel-based global optimization algorithms such as classical BP and Graph Cut(GC) algorithm, this algorithm can greatly reduce the error matching rates especially in textureless regions and occluded regions.
出处
《计算机工程》
CAS
CSCD
2013年第7期257-260,278,共5页
Computer Engineering
基金
天津市教委科技发展基金资助项目(20090718)
关键词
立体匹配
视差
均值偏移
图像区域分割
平面拟合
置信传播
stereo matching
disparity
mean shift
image region segmentation
plane fitting
Belief Propagation(BP)