Research On Face Recognition Algorithm Based On GA-BP To Eliminate The Influence Of Local Occlusion

Face recognition is one of the most widely used applications of artificial intelligence.As far as the development of science and technology is concerned,face recognition has been used in the life of mobile phone unlocking,security payment and so on.However,these technologies are all built in the absence of occlusion.When faces are obscured by sunglasses and scarves,we often need to remove the accessories to be identified.In view of this problem,the following work is done in this paper.1.Preliminarily identified according to the organ characteristics.First,the features of the eye and mouth organ were extracted.The eyes and the mouth have good edge features.This paper uses gray integral to locate its specific location.And the extracted organ features are input to the BP neural network for training and learning.However,because the image is transformed from two dimensions to one dimension and the dimension is higher,the network will produce time redundancy and complex structure during the training process.Therefore,the PCA is used to reduce the dimension in the input,and the data is preprocessed.2.Genetic algorithm(GA)is used to optimize the BP neural network.In theory,the main purpose of the BP neural network is to train the error between the target and the expectation.The gradient descent method is used to adjust the weights and thresholds of the network neurons.The main disadvantage of the BP algorithm is that the weight and threshold of training converge to the local minimum,and the global optimal is not guaranteed.In addition,the BP only adjusts the given weight and threshold to minimize the output error.The genetic algorithm is a very powerful adaptive global optimization algorithm because of its implicit parallelism.Using GA to optimize the weight and threshold of BP can improve the shortcoming of the local optimal solution,improve the convergence speed,and help to improve the face recognition rate.3.Image fusion in RGB space.R,G,and B are three visual colors that do not use frequency,and contain more abundant information.In this paper,the face reconstruction based on R,G and B is proposed.The reconstructed face is more effective in removing the shade of the sunglasses and scarves.When the R,G and B components are selected,the image fusion of the three components using PCA can be used more effectively and more completely to restore the obscured part of the image to be measured.The experimental results show that the fusion based on RGB space not only effectively removes the occlusion,but also improves the matching speed and recognition rate of the face to a great extent.4.Face global feature extraction.LDA is the projection of the high dimensional data to the feature space,and in the subspace,a set of projection vectors of the largest interclass scatter matrix and the minimum subclass scatter matrix are determined.That is to say,the pattern has the best separability in the feature space.However,LDA is always plagued by the small sample(SSS)problem.Therefore,the use of PCA to reduce its dimension makes S_w nonsingular before the use of LDA.In the end,the K-Nearest Neighbor(KNN)is used for classification and recognition.Finally,the research method is carried out in the AR face database under different conditions and is compared with the existing methods.Experiments show that this method can remove face occlusion more effectively and achieve better face reconstruction,which effectively improves the face recognition rate under sunglasses and scarves.