基于局部分阶段归一化的复杂光照人脸识别

Face Recognition under Complex Illumination Conditions based on Localized Hierarchical Normalization

以往算法通过低通、高通滤波的模式实施人脸图像的光照、光照不变量的估计,存在邻域像素信息混杂的问题,无法准确地从人脸图像中提取人脸本征.考虑邻域像素成像光照的相关性,提出基于邻域像素成像光照消除的局部分阶段归一化光照不变特征提取模型,可获取人脸图像的多细节光照不变特征,并通过内积度量构建基于多层次匹配的类别鉴别方法.Yale B和Extended Yale B人脸库上的试验结果表明,该算法能明显提高复杂光照人脸识别性能,明显优于当前先进算法.

The existing algorithms using low-pass and high-pass filtering modes for estimating illumination and illumination invariants from a face image, often cause the confusion of the neighborhood pixel information, and the illumination invariants. Based on the correlation of adjacent pixel imaging illumination, a local hierarchicalnormalization model for extracting illumination invariants is proposed, which can obtain multi-level illumination invariant features of face images. In addition, a discrimination method based on multi-level matching is constructed by adopting inner product measure. The experimental results on the Yale B and the Extended Yale B face database show that the proposed method can improve the performances of face recognition under complex illuminatithan current advanced algorithms.