基于群正则化相似性度量的人脸识别方法

Face Recognition Based on Cohort Normalization Similarity

安全监控应用中,受光照、阴影和运动模糊等影响,通过人脸检测算子检测到的图像可能包含不完整的人像信息,严重影响到识别的精度。提出一种人脸选择算法,从给定的候选人像集合中选择一个高质量人像的子集,然后应用基于集合的人像识别算法进行识别,有效地提高识别的精度。在公开的人脸识别数据库Honda/UCSD和Choke Point的实验结果显示,使用子集选择的算法能明显提高现有基于集合的人像识别算法的精度。

In surveillance applications, face images captured with different illumination, shadowing, and motion blur over the sequence, the snapshot may contain non-face or incomplete face component. Addresses the problem of face recognition with an image set-based approach.The proposed method is more robust. It doesn＇t need an alignment of the face. It automatically selects high-quality images for face recognition during testing and training. Experimental results on the shared video database Honda/UCSD and Choke Point show that the proposed framework method has been promising potential for use in the image set-based automatic face recognition applications.