Multi-pose Face Detection Method Based On SVM

Research on multi-pose face detection has great significance to face detection toward the practical application, but also by the general concern of the field of pattern recognition. The low accuracy of multi-pose face detection in recent years, is a more difficult to break through the technical problems. Support vector machine is built on the VC dimension theory and structural risk minimization principle on the basis of statistical learning theory, which based on a limited sample information between the complexity of the model and ability to learn to seek the best strategy in order to obtain the best promotion ability.lt can better solve the complex problem of face detection in non-linear, small sample size, the local minimum points. The support vector machine used in face detection, has a great significance to face detection technology.In this paper, it study multi-pose face detection algorithm in-depth, which combined with hyper-sphere SVM technology, and introduces the hypersphere support vector machines based on multi-pose face detection, this paper has a detailed analysis on face detection and domestic and foreign and application of methods, which given the collaborative model of face detection, and introduced data preprocessing method which can improve the overall performance of detection, Final It has a details of the hypersphere-based human face detection algorithm model. The main research work includes:(1) It analysis the current domestic and international multi-pose face detection research background and research status of multi-pose face detection method based on support vector machine, the advantages and disadvantages of the common multi-pose face detection range of applications and related technologies are analyzed.(2) For an image large areas of the non-human face, this paper presents a multi-pose face detection collaborative model. The model consists of a group composed of hyper-sphere support vector machines, they were divided into three layers:the first layer has a detector, the second layer has three SVM, the third layer has nine SVM, a total of13SVM. These SVM is sophisticated detection design by drill, collaborative complete face detection tasks. Three-layer model has been designed with two advantages:one can accelerate the speed of face detection, on the other hand improve the detection of targeted, making more sophisticated layer by layer to fulfill local area detection.(3) This paper improved k nearest neighbor (KNN) algorithm and compared the ordinary KNN algorithm and improved KNN algorithm advantages and disadvantages. The improved KNN algorithm for the detection of supersphere overlapping samples, improved face detection accuracy.(4) It designed and implemented based on hypersphere multi-pose face detection algorithm. Constructed based on hypersphere multi-pose face detection, analysis hypersphere support vector machines for multi-pose face detection advantages and disadvantages.(5) This paper carry out the relevant experiments based on the proposed supersphere multi-pose face detection technology, and the experimental results are analyzed in detail, as opposed to the traditional SVM-based face detection, face detection proposed algorithm accuracy has improved; through layer by layer filtration methods to improve the detection rate of face detection, face detection has a positive effect towards practical applications.This multi-layer detection model based on given hypersphere support vector machines based on multi-pose face detection method, which binds KNN algorithm to handle overlapping regions hypersphere samples overlap within the sphere increases the accuracy of the sample. Model for the cooperative performance of the relevant experimental results show that the method has good detection results, the overall performance is relatively highFinally, research and design work are summarized, and pointed out the support vector machine multi-pose face detection algorithm direction for further work.