Study On Multi-pose Bionic Face Detection Based On AP And Invariance Features

The development of bionic algorithms leads combination various methods to be a trendfor the study of face detection. For the key issue of what kinds of methods that can be used,this thesis deeply studies how to improve the correct detection rate of multi-pose faces byconsidering the advantages as well as disadvantages of the existing face detection algorithms.The main work of this thesis is as follows,1. According to the problems that the general SVM (Support Vector Machine) neuralnetwork selects training sample sets randomly,which affects the precision of classifier, andthe PCA (Principal Component Analysis) algorithm has some limitations when there existsmultiple correlation among the original features, the AP (Affinity Propagation) clusteringalgorithm is applied to select the training sample set, which can remove redundantinformation of sample set and overcome the limitations of PCA simultaneously. Experimentalresults show that SVM classifier based on PCA and AP clustering can decrease the trainingand testing time and has high face detection rate and better non-face rejection rate comparedwith the traditional methods to some extent.2. A multi-pose face detection algorithm is realized by integrating different detectionmethods. For a face image to be detected, after preprocessing, the whole image window issearched based on template matching and evolutionary Agent algorithm. Then the candidateregions of faces to be detected can be located preliminary. Finally the real facial regions canbe determined and marked using the trained SVM classifier designed by AP and PCA.Experimental results show that the proposed method has high correct detection rate and lowfalse detection rate.