摘要
针对在立体匹配过程中的低纹理和视差不连续的问题,提出了一种基于图像分割的立体匹配算法。首先采用Mean-Shift算法对彩色图像进行图像分割,并认为分割区域块内的像素视差是平滑的,然后采用较大窗口匹配方法,提取左右彩色图像相似像素点作为种子点,根据区域内的平滑约束条件,利用小窗口匹配方法将种子点向周围区域扩散,最终得到稠密视差图。实验结果表明该算法相比传统的自适应匹配算法,视差不连续区域匹配误差降低10%左右。
For the problem of the low texture region and disparity discontinuity in stereo matching, a stereo matching algorithm based on image seg- mentation is presented. Firstly, color image is segment by applying mean shift method. And assuming that the disparity of pix in the segment is smooth. Then pixels which are similar in right and left color imageare extractedby using large window as seed points,and then spreading seed points to the surrounding with the area of smooth constraint conditions by small window. Finally, the disparity map is generated. The experimental results show that the mismatching of algorithmis about ten percent lower than the traditional adaptive matching algorithm in disparity discontinuity region.
出处
《电视技术》
北大核心
2014年第19期5-7,12,共4页
Video Engineering
基金
国家"863"重大专项(2012AA03A301
2013AA030601)
国家自然科学基金项目(61101169
61106053)
福建省自然科学基金项目(2011J01347)
关键词
立体匹配
遮挡
视差
图像分割
stereo match
occlude
disparity
image segmentation
