基于改进Census变换的立体匹配算法

针对基于传统Census变换的立体匹配方法精度不高的问题, 提出一种基于坐标模板Census变换的立体匹配方法。该方法首先设计一个坐标为高斯分布的变换模板指导Census变换; 其次, 以Census变换结果之间的Hamming距离作为匹配代价求取初始视差, 同时使用SSE4. 2指令提高Hamming距离计算效率; 再次, 使用非局部代价聚合方法优化视差结果; 最后, 获得精度较高的视差图像。实验结果表明, 该算法具有较好的匹配精度和较高的匹配效率。

Improved Census transform for stereo matching

In order to improve the accuracy of stereo matching based on Census transform, this paper proposed an improved Census transform. Firstly, it designed a transform template whose coordinated follow Gaussian distribution as a guidance of Census transform; secondly, it used the Hamming distance as the matching cost for computing initial disparity, meanwhile, accelerated the speed of Hamming distance computing by employing SSE4. 2 instructions; lastly, it applied a non-local cost aggregation method to refine the disparity. Several experimental results demonstrate that this method is more accuracy and efficiency.