复杂光照下的视频人脸检测与识别研究

Study of Video Face Detection and Recognition under Complex Illumination

为了解决复杂光照下的视频人脸检测与识别率受影响的问题,提出了一种光照不变的人脸检测与识别方法。该方法基于Retinex理论,提取光照不变分量,然后用于训练AdaBoost分类器;对输入的视频序列也进行相同的光照预处理,然后用训练的AdaBoost分类器进行人脸检测;把检测到的光照不变人脸图像采用分块加权LBP进行特征提取,采用欧氏距离与最近邻分类器进行分类。实验结果表明:该方法能有效提高视频人脸检测率与人脸识别率,而且对于人脸检测与识别只需要一次光照处理,具有更高的效率。

In order to solve video face detection and recognition rate affected by the problem of complex light, an illumination invariant method of face detection and recognition is proposed in this paper. The method is based on the Retinex theory. Firstly, the illumination in- variant component is extracted from input image. Then it＇s used to train the AdaBoost classifier. Illumination invariant component is also extracted from input video sequence to detect face by using the trained AdaBoost classifier. Block weighted LBP algorithm is used to ex- tract illumination invariant face feature. Euclidean distance and the nearest neighbor classifier are used to classify the feature vectors. The experimental results show that the method can effectively increase the rate of video face detection and face recognition, and it only needs an illumination pretreatment for video face detection and recognition. Therefore, it has higher efficiency.