The Influence Mechanism Of Gabor Wavelet Parameters On Facial Feature

With the rapid development of artificial intelligence,the extraction and classification recognition of facial feature has become a research hotspot.In this thesis,the influence mechanism of Gabor wavelet parameters on facial feature is studied.The main innovation points are as follows.(1)Aiming at the problem that each decision tree has the same decision-making ability in the traditional Random Forests algorithm,which is not conducive to the final classification,the Weighted Random Forest algorithm is proposed to locate the key points of facial features.Firstly,the face parts are detected based on Haar feature and AadBoost algorithm,and then the key points of facial feature are located.Finally,the collected face images are preprocessed.(2)Aiming at the influence of 2D-Gabor wavelet parameter change on face characteristics,the relationship corresponding tables among the changes of 3 parameters:Gabor filter direction ?,filter frequency f and root-mean-square deviation cr.The experimental results show that the detail texture features of the face image become blurred when the face characteristic shifts from the high frequency characteristic to the low frequency characteristic conversion.(3)Aiming at the problem that Gabor wavelet extract facial feature with high dimension and redundant f-acial feature information,this thesis proposes Gabor multi-scale mix feature algorithm in 8 directions(Gabormix8)to extract facial feature.Thereby facial feature information is enhanced,unnecessary feature information is eliminated and the Gabor feature dimension from 103040 to 20608.(4)Aiming at the problem that BP neural network is easy to fall into local optimal value and slow convergence speed in solving the optimal problem,this thesis proposes Artificial Fish Swaram optimization BP neural Network(AFSA-BPNN)classification recognition algorithm.By adjusting the parameters of Artificial Fish Swarm Algorithm and simulating the movement laws of natural fish,the fish can selectively foraging,flocking and following behavior,and realize the global optimization.In this thesis,a face recognition system is constructed by using VS2013 development platform and OpenCV 2.4.8 function library,experimental results show that the facial features extracted by the Gabormix8 algorithm are identified under BPNN classifier,the average recognition accuracy can reach 92.23%.With 0.01 as the error accuracy,the AFSA-BPNN training steps improved by 35.44%compared with BPNN training steps.At the same time,the facial features extracted by the Gabormix8 algorithm are identified under the AFSA-BPNN classifier,the average recognition accuracy can reach 94.21%,which improved by 9.22%and 1.98%compared with KNN and BPNN classifiers.