Research On Multi-view Face Detection Method In The Wild

Face analysis and understanding in wild is an important research problem in the field of computer vision and pattern recognition.It has wide application prospect and huge market demand in video field retrieval,intelligent video surveillance,human-computer interaction and intelligent security etc.Face detection is the core technology and application foundation of face analysis.However,it has become a challenging research topic due to external factors such as angle of view,deformation,illumination and occlusion.This paper focus on the multi-view face detection algorithm based on the multi-channel feature to solve the problem that the change angle of face in wild.This method can effectively suppress the noise information in the background of the image and improve the accuracy and speed of face detection.The main work of this paper is as follows:(1)A kind of discriminative weight pooling feature for multi-view face detection is proposed to solve the problem that the aggregate channel feature of face in wild has a low detection rate which affected by many kinds of complex factors.Based on the multi channel map feature,this method introduces the middle layer rectangular filter with strong discriminant,which improves the discriminant ability of the feature.The filter is based on the average face shape statistics,and uses linear discriminant analysis and its improved imbalance embedding algorithm to study the distribution information of positive and negative training samples.The validity of the method is verified by comparing with the original multi-channel feature and some state-of-art algorithms on the FDDB database.(2)A multi-view face detection algorithm based on multi-channel maps discriminative projection HAAR feature(PHF)is proposed to solve the problem that the traditional HAAR feature has low detection rate and the distribution of positive and negative training samples is imbalanced.Firstly,the underlying multi-channel maps of the face training samples is calculated.Then,the underlying ACF feature map is filtered by the linearly discriminant projection learning enhanced HAAR feature based on the positive and negative training samples.The asymmetric AdaBoost algorithm is used to select the strong discriminant PHF feature,which effectively suppresses the imbalance of positive and negative sample space.The experimental results show that this method has higher detection accuracy and faster detection speed than the state-of-art methods.(3)A multi-view face detection algorithm based on local module dictionary of face is proposed due to the relative stability of some local structure of face under different face angles.Firstly,the local structure of face is represented by the histogram of oriented gradient(HOG)feature.Then,the significant facial local module with similar semantic categories are obtained by clustering algorithm,and the support vector machine is used to train multiple classifier of different face local module.In the detection phase,the model classifier detection results are handled by Hough Voting to obtain the face position.The experimental results show that the method can effectively improve the detection rate.