摘要
为了在消除弱纹理区域匹配歧义性的同时保证纹理区域的匹配率,提出了一种基于弱纹理区域检测及视差图融合的立体匹配算法。该算法首先根据输入图像的颜色(灰度)变化情况检测出弱纹理区域,然后基于这一检测结果,对输入图像对应用改进的极线距离变换算法,以提高弱纹理区域像素的可区分性,接着,采用窗口匹配算法和置信度传播算法分别对原始输入图像和极线距离变换后的图像计算视差图谱,最后,以弱纹理检测的结果为基准,对这2张视差图谱进行融合,以实现在弱纹理区域和纹理区域的同步最优匹配。通过对弱纹理化后的Middlebury图像库中图像的实验表明,在几乎不增加计算复杂度的同时,该算法的匹配率比当前先进算法提高至少20%,同时,实验还表明了该算法对照度不一致输入图像对匹配的鲁棒性。
In order to eliminate the ambiguity of low-texture region matching, while preserve the matching rate of the texture regions, a novel stereo matching algorithm based on low-texture region detection and disparity map fusion is proposed. The procedure is proceed in the following way：first of all, the low texture region is detected based on the color or the grayscale of the input image ; based on this result the improved epipolar distance transformation algorithm is used on the input image pair to improve the distinguishing ability of the pixels in the low-texture regions ; and then the window matching algorithm and belief propagation algorithm are adopted to respectively calculate the disparity map for the original input image and the image after the improved epipolar distance transformation. At last, the initial disparity maps calculated from the original and the transformed image pairs are fused according to the detection re- suits of the low-texture regions;and the synchronous optimal matching of the low-texture regions and the texture re- gions is achieved. The experiments on the low-textured images in the Middlebury image database indicate that the proposed algorithm improves the matching rate by at least 20% compared with the state of art algorithms, almost with- out increasing the calculation complexity;moreover, the algorithm also has robustness to the matching of the input im- age pair with luminance variation.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第9期2074-2079,共6页
Chinese Journal of Scientific Instrument
基金
国家863计划(2010AA7080302)资助项目
关键词
图像处理
立体匹配
弱纹理
极线距离变换
image processing
stereo matching
low-texture, epipolar distance transformation
