基于相位重构的光照人脸识别

Face recognition under varying illumination based on phase reconstruction

光照是人脸识别中的一个难题。信号经傅立叶变换后相位谱包含丰富的纹理结构信息,去除幅度谱,对图像进行单纯相位谱重构可以有效地减少光照影响。对角化主成份分析(Diagonal PCA)方法是对主成份分析(PCA)方法的有效改进,其计算量更小识别率更高。采用相位重构图像作为训练样本进行对角化主成份分析来解决光照问题,取得了较为理想的识别效果。

The face recognition under varying illumination is still a challenge. In the Fourier representation of the signals the magnitude and phase tend to play different roles and in .some situations the phase contains more illumination information. It is proved that the diagonal PCA is more accurate than the original PCA and it needs less computation. The phase spectrum is combined with the diagonal PCA to solve the problem without performing any pre-processing. The experiment shows that the proposed method can achieve very encouraging performance in varying illumination conditions.