Study On Face Detection Based On Method Combined RVM With SVM

Face detection in complex background is widely used in face recognition,video retrieval,human-computer interaction and so on.It is one of the hot spots in the field of computer vision and pattern recognition.The existence of the pose,illumination,noise and occlusion makes the face detection in engineering design difficult.It's challenging to quickly and accurately detect the faces.Support Vector Machine(SVM)based on statistical learning theory,can effectively solve the small sample learning problem.However,in practice,SVM has many disadvantages,such as high computational complexity,many support vectors,slow training speed and so on.These shortcomings also lead to the detection time of pure SVM face detection system is very long.And the Relevance Vector Machine(RVM)based on Bayesian framework can make up for these deficiencies.The main work is to study the face detection algorithms.It is found that the detection rate of such algorithms is high with the slow detection speed after the analysis and summary of the face detection based on SVM.To solve this problem,this paper proposes a face detection based on method combined RVM with SVM.The face detection system is composed of three classifiers,and the simplest average face template matching is used as the first level classifier,which can quickly filter out a large number of simple background windows,and greatly increase the detection speed.The RVM as second classifier has the higher precision.It can filter out almost all of the background windows for assuring all face windows pass.While the nonlinear SVM with the highest precision confirms as long as a small windows.This coarse-to-fine design can make full use of the high speed of average face template matching,the high accuracy of SVM,as well as the higher speed and accuracy of RVM.The MATLAB simulation results show that the first two level classifiers filter out nearly 99% of the background windows with only about 1% of the windows processed by the SVM classifier.The algorithm not only greatly increases the detection speed,but also reduces the number of false detections and improvs the detection rate.