改进的双目立体视觉正面脸合成

Frontal Face Synthesis Based on Improved Binocular Stereo Vision

为了解决人脸在不同姿态下难以识别的问题,提出了一种基于改进的双目立体视觉的正面人脸合成算法.针对人脸图像纹理平滑性对特征匹配带来的困难,设计了一套新的匹配算法,获得了更准确的人脸视差图.根据视差图求出人脸的三维坐标,参照人眼特征点定位结果求出人脸偏转角度,并将人脸合成为正面图像.实验结果表明,该方法比其他方法具有更高的真实性和准确性,同时具有较快的处理速度.

To improve the face recognition performance under different poses, we propose an improved frontal face synthesis method based on binocular stereo vision, which can effectively and quickly synthesize a frontal face. A modified matching algorithm is designed to deal with smoothness of faces. Then 3-D coordinates are generated according to the disparity image, and the pose angle of the face calculated according to the location of eyes. A frontal face is generated according to the pose angle using the method of binocular stereo vision. Experimental results show that the method is precise, practical, and fast.