Research On Face Recognition Based On Multi-Feature Fusion

With the continuous development of intelligent technology,the focus of research is how to identify the personal identity quickly,safely,and effectively.As a biological feature that everyone has,face has features such as convenience,uniqueness,and security,and it has been widely used in various fields.An improved multi-feature face detection algorithm is presented by studying the research of face detection based on the skin color feature and the method based on AdaBoost.This method filters most of the background regions in the image and extracts the area of interest by the skin color detection based on regional feature fusion.Then,to obtain the likeness human face area,extracting the Haar-Like integrogram features and detecting by the AdaBoost cascade classifier.Finally,using the binocular positioning which can not only judge the face-like images again,but also use the empirical values of the geometric features of the eyes and face contours to obtain the accurate face image.The experimental results show that this method can improve the detection speed.Aiming to the problem that the LBP operator can not describe face feature full-scale,a multi-scale layered LBP fusion feature method is presented.Firstly,extracting face texture features by using LBP operators of different scales,and then using the layering principle to obtain the layered LBP fusion feature,next cascading the layered fusion feature of different scales to describe the different scale texture of face.Using the PCA-LDA feature reduction method to solve the problem of excessive dimension.Experimental results show that the using of fusion feature can effectively improve the face recognition rate.In addition,considering the disadvantages of traditional KNN classifiers,which has high computational complexity and insufficient utilization of sample differences,an improved KNN classifier is proposed.Experimental results show that the classifier improved has better effect of classification.The face recognition system constructing by using the presenting algorithm in this paper can achieve real-time detection of face in the video,and complete the matching recognition by comparing the improved fusion feature with the face data in the database.It is of great significance for identification.