基于最小扭曲变换的正面人脸图像合成

Multi-view Face Synthesis Using Minimum Bending Deformation

摘要: 二维正面人脸图像在视频监控、人脸识别等领域有重要的实际意义,为此提出一种利用多幅非正面人脸图像自动合成正面人脸图像的算法.从多幅非正面人脸图像中选择扭曲形变最小的像素进行合成,并使用加权平均来平滑合成图像,从而使得合成的正面人脸图像尽可能自然.实验结果表明,该算法所合成的正面人脸图像可以十分有效地逼近真实正面人脸图像,在人脸识别中也可以有效地保持人脸图像的身份特征不变.

Abstract: 2D frontal view face is significantly important in video surveillance and face recognition.In this paper,an algorithm of frontal view face synthesized from multi-view face images is proposed.The pixels of synthesis image are chosen from multi-view face images based on minimum bending deformation and smoothed by weighted average.Experiments demonstrate that the synthesized frontal view face could approximate the ground truth frontal view face and preserve the identification information for face recognition.