基于引导滤波及视差图融合的立体匹配

Stereo matching based on guided filtering and disparity map fusion

摘要:

立体匹配是双目视觉领域的重要研究方向。为在保证图片纹理区域匹配精度的同时降低弱纹理区域的误匹配率，本文在传统引导滤波的基础上提出一种基于引导滤波及视差图融合的立体匹配方法。首先，根据图像颜色相似性将图片划分纹理较丰富区域和弱纹理区域。接着，分别采用不同参数的引导滤波进行代价聚合及视差计算得到两张视差图。然后依据纹理区域划分的结果对获得的两张视差图进行融合。最后，通过左右一致性检测、加权中值滤波等视差优化步骤得到最终视差图。通过对Middlebury测试平台上标准图像对的实验表明，该方法在6组弱纹理图像上的平均误匹配率为9.67%，较传统引导滤波立体匹配算法具有更高的匹配精度

Abstract:

Stereo matching is an important research direction in the field of binocular vision. In order to ensure the matching accuracy of image texture regions and reduce the mismatch rate of weak texture regions, this paper proposes a stereo matching method based on guided filtering and disparity map fusion based on traditional guided filtering. First, the image is divided into areas with rich texture and areas with weak texture according to the similarity of image colors. Then, two disparity maps are obtained by cost aggregation and disparity calculation using guided filtering with different parameters. Then, the two disparity maps obtained are merged according to the result of texture region division. Finally, the final disparity map is obtained through disparity optimization steps such as left and right consistency detection and weighted median filtering. Experiments on standard image pairs on the Middlebury test platform show that the average mismatch rate of this method on 6 sets of weak texture images is 9.67%, which has a higher matching accuracy than the traditional guided filter stereo matching algorithm.