采用邻域差值特征模板的立体匹配方法

提出了一种基于邻域差值特征模板的立体匹配方法。该方法通过计算一幅图像中间隔一定距离的两点的像素邻域差值作为特征模板，再计算另一幅图像中在同一扫描线上间隔同样距离的像素邻域的差值，将它与特征模板的偏差作为匹配标准，偏差最小的点就认为是匹配点。在同等条件下，与区域相关法相比，该方法能够将运算的速度提高3倍，且同样能够得到精确浓密的视差图。实验证明，该算法结构简单，易于实现，能够处理复杂的场景，具有良好的实验效果和实用价值。

Stereo matching algorithm based on feature template with neighborhood difference

A stereo matching algorithm based on feature template with neighborhood difference is proposed in the paper. In left image, the difference of the two neighborhoods whose centers are two points on the same scan line and have a fixed interval has been calculated and regarded as the feature template. Then in right image, the difference of the two neighborhoods whose centers are two points has been obtained too, these points have the same scan line and interval as the corresponding points in the left image. Then the warp between the feature template and the corresponding difference which is obtained from the right image has been reckoned. The point which has the minimum warp is the best matching point. Compared with area-related method for the same case, this method can improve the operation speed by a factor of 3. Experimental results show that this algorithm is effective and it can be easily implemented for processing complex scene.