顾及图像分割信息的半全局立体匹配算法研究

Research on semi-global stereo matching algorithm considering image segmentation information

针对传统半全局立体匹配算法(SGM)在高分辨率图像弱纹理以及视差不连续区域的误匹配问题,提出了一种顾及图像分割信息的SGM算法。该算法在代价计算阶段,首先根据图像分割信息自适应调整匹配窗口大小,采用不同状态信息的改进Census变换计算初始代价,解决传统算法对Census变换窗口中心像素依赖的同时减少了匹配时间;在代价聚合阶段,将图像分割信息与传统SGM算法的全局能量函数进行有机结合,提高算法在弱纹理以及深度不连续区域的匹配效果;最后通过左右一致性检测和子像素细化得到优化后的视差图。所提算法利用Middlebury平台标准数据进行验证,实验结果表明,平均误匹配率为4.54%,与传统SGM算法和一些改进算法相比,该算法能够在影像弱纹理和视差不连续区域获得更高匹配正确率。

Aiming at the mismatching problem of traditional semi-global matching(SGM)in high-resolution images with weak textures and disparity discontinuities,an SGM algorithm that takes into account the image segmentation information is proposed.In the cost calculation stage of this algorithm,the size of the matching window is first adaptively adjusted according to the image segmentation information,and the improved Census transform with different state information is used to calculate the initial cost,which solves the traditional algorithm′s dependence on the center pixel of the Census transform window and reduces the matching time.In the cost aggregation stage,the image segmentation information is organically combined with the global energy function of the traditional SGM algorithm to improve the matching effect of the algorithm in weak texture and depth discontinuous regions.Finally,the optimized disparity map is obtained through left-right consistency detection and sub-pixel refinement.The proposed algorithm is verified by using the standard data of the Middlebury platform.The experimental results show that the average mismatch rate is 4.54%.Compared with the traditional SGM algorithm and some improved algorithms,the proposed algorithm can obtain higher matching accuracy in the weak texture and discontinuous disparity areas of the image.