针对当前视差图生成算法中常用的匹配代价对光晕等复杂光学畸变比较敏感的问题,在对相机成像模型进行分析的基础上提出了一种基于张量分析的匹配代价。和目前常用的匹配代价相比,该匹配代价不但对光晕效应具有更强的鲁棒性而且能够更加准确的反映图像的局部结构信息,有效降低了匹配的歧义性。在本文中,张量在积分图像上构造,张量间的距离在黎曼流形上测量;此外,提出了一种简单有效的视差图后续处理方法进一步修正了初始视差图中的不可信视差。最后,在光照变化和光晕效应等非理想情况下将本文方法与其他几种较新的视差图生成方法做了比较,实验结果表明本文方法具有更强的鲁棒性和更高的视差估计精度。

To reduce the sensitivity of the matching costs to complex optical distortions, such as vignetting, by fully exploiting the imaging model, a new matching cost using tensor analysis has been put forward. Compared with conventional costs, the proposed cost is resistive to complex distortions such as lens vignetting. Besides, it can reduce the matching ambiguity by reflecting the local image structure to a finer degree. We construct the proposed tensors rapidly on an integral image and measure their distance on the Riemannian manifold. In addition, a simple but effective disparity post-processing method is employed to further emend errors in the initial disparity map. Finally, the proposed method is compared with several other state-of-the-art disparity map generation methods under some challenging situations such as illumination variations and vignetting, the experimental results show that the proposed method is more robust and has higher disparity estimation accuracy.