Research On Face Detection Algorithm Based On Machine Learning

Face detection is an important part of image processing,and it is an important research field of computer vision.It has been widely used in intelligent security,humancomputer interaction and confidential authentication.At present,the face detection methods have limitations in terms of various angles,complex background,and illumination changes.This paper explores the face detection methods in the framework of machine learning theory,and proposes two face detection algorithms to improve the accuracy and robustness:1.In order to improve the accuracy of face detection under varying views,this paper proposes a multi-view face detection algorithm based on BING and multi-texture CS-LBP features.In this algorithm,Binarized Normed Gradients(BING)is introduced,and twostage models are trained to achieve the primary selection of face regions.This paper also proposes multi-texture central symmetric local binary pattern(CS-LBP),which expresses the face image from four texture directions respectively,which effectively enhances the recognition ability of the feature.In addition,by calculating the facial topological distance to complete the division of face pose perspective.A face detection structure combining multiple single-view nesting-structured cascade classifiers and multi-layer perceptron(MLP)cascade is proposed,which improves the robustness of the algorithm to multi-view face detection.Experiments on FDDB and PASCAL Faces show that the proposed face detection algorithm can effectively improve the accuracy and speed of face detection.2.In order to make up for the limitation of single feature in face description,this paper proposes a face detection algorithm based on localized multiple kernel learning feature fusion.The algorithm combines multi-texture CS-LBP features,ORB and HOG,and expresses face images from three aspects: texture information,facial landmark extraction and gradient characteristics.In this paper,each feature is sparsely coded,and the spatial pyramid matching visual word bag model is used to complete statistical expression of coding vectors.It improves the capacity of resisting disturbance of the feature.In order to realize the effective combination of multiple features,this paper proposes a feature fusion algorithm based on localized multiple kernel learning(LMKL).It introduces a multi-feature fusion method and proposes a new gating function.By training,each feature is given a corresponding weight to ensure the rationality of weight distribution.Experiments show that the face detection algorithm proposed in this part can reflect the characteristics of face images more comprehensively and effectively improve the robustness of face detection.