基于极线距离变换的人脸立体匹配算法

Face stereo matching using epipolar distance transform

针对人脸立体匹配中低纹理区域的误匹配率较高的问题,将极线距离变换应用到人脸立体匹配当中。这种变换将图像中像素点的灰度值转化为极线方向上相应区域的位置,使得在低纹理区域中灰度值相似的像素点变得容易区分。因此可以用于提高立体匹配算法在低纹理区域的匹配精度。但是在应用到人脸立体匹配过程中会出现高纹理区域的鲁棒性和精度下降的问题,通过对不同极线长度的极线距离变换结果提取有利的部分进行匹配来解决这一问题。实验结果表明,采用改进后的方法,在人脸的低纹理区域仍然能获得较高的匹配精度,并且降低了计算的复杂度,同时对于人脸中的高纹理区域具有较好的鲁棒性。

To address the problem of high error rate in matching human face＇s low-texture regions,a stereo matching algorithm based on epipolar distance transform is presented.Image intensity values are converted to a relative location along the epipolar line,so pixels in the low-texture regions become distinguishable,which improve stereo matching algorithm in matching precision of low-texture regions.When the method is applied to face high texture,the robustness and accuracy of stereo matching will decline.To solve the problem,favorable parts of epipolar distance transform results are extracted in different epipolar length.The results of experiments show that higher matching accuracy and lower computational complexity can be achieved in stereo matching of human face with this transformation,and the robustness of the transform for high-texture regions is demonstrated.