基于分级变换的自适应立体对应算法

通过分级变换将图像从灰度空间转换到新的等级空间,然后构造相应的匹配代价函数计算两个图像点之间的最大相似度,从而找出对应点和偏移值。分级变换可以有效的解决在立体对应中经常遇到的图像噪声、失真及左右图像的亮度差异等问题。大多数的自适应立体对应算法是以偏移量和灰度值两个自变量来构造代价函数,而构造合适的代价函数是一个困难的问题。本文中提出自适应窗选择算法只与灰度值有关。首先通过边缘检测提取出灰度边缘信息。本算法仅根据灰度边缘信息就可以进行自适应窗的选择。自适应选择图像窗的过程与偏移值无关,从而降低了构造代价函数的难度。实验结果说明本算法能够生成准确度较高的深度图,是一种较好的局部立体对应算法。

Rank Transform Stereo Algorithm with Adaptive Window

The stereo algorithm is composed of an improved rank transform method and a new adaptive window method. The improved rank method transforms image data from gray-level plane to the rank level plane. It could solve the difficult problems for area-based stereo algorithm such as image distortion, noise and brightness difference etc. Most adaptive stereo matching algorithms utilize an evaluation function of disparity and intensity to choose the appropriate windows for image pixels. To construct an appropriate evaluation function is a very difficult problem. While our algorithm chooses the appropriate window only by the intensity variance based on edge detection approach and the process has no relation with disparities. It reduces the complexity to construct the evaluation function. To demonstrate the effectiveness of the algorithm, it has been tested by stereo images with ground truth in Middlebury stereo database. The results show that our algorithm has high correct matching rate and it is comparable to the best local algorithm. Meanwhile it is robust with few parameters to be set.