Research On Key Technology For Face Recognition Under Complex Illumination

Face recognition is widely used in the fields of economy,military,security,criminal investigation and so on due to its non-contact characteristics.However,because it is easily affected by illumination,expression,posture,etc.,a very robust face recognition system needs to be strengthened,and the task of face recognition research has a long way to go.Among them,the change of lighting conditions has a more prominent impact on face recognition.To apply face recognition in more occasions,it is necessary to solve the lighting problem.This article studies the impact of lighting on face recognition.The specific analysis content is as follows:1.Study the effects of three normalization algorithms,logarithmic transformation,histogram equalization,and gamma transformation,on the image quality of human faces under complex lighting conditions,and then verify and analyze the two invariant features of MSR and self-quotient images.Extract the algorithm and conduct experiments on the YaleB and CMU-PIE face databases to find the differences between different algorithms in dealing with face images under complex lighting conditions.2.The gradient face algorithm greatly improves the face image of complex lighting.In order to obtain better lighting-invariant features of face images,based on the gradient face algorithm,this paper proposes a multi-directional gradient face algorithm,which calculates the weighted fusion of the gradient features of six directions under the lighting conditions of the face Get multi-directional gradient face images.Afterwards,in order to perform better feature selection and extraction,the obtained multi-directional gradient face is combined with several typical feature extraction algorithms to select representative features.The experimental results show that the face image is processed by multidirectional gradient face algorithm and then feature extraction can significantly improve the efficiency of the recognition system.3.The Retinex algorithm can solve the problem of recognition of complex lighting face images well.It is a classic algorithm to solve the lighting problem,but it will produce halo artifacts under the dramatic change of lighting.In view of the phenomenon of halo artifacts in the Retinex algorithm,in order to improve the quality of face image processing under complex lighting conditions by the Retinex algorithm,a Wiener filtering method is proposed to estimate the illumination,which can get the illumination of the face image well Unchanging features,eliminating halo artifacts.The experiment is based on subjective visual effects and objective experimental data.By comparing the processing effects of several algorithms,the processing ability of this algorithm and its effectiveness in solving complex illuminated face images are proved.4.The face image under complex lighting conditions is processed by the optimized Retinex algorithm to obtain the feature of invariant lighting components.For the subsequent research of face recognition,feature fusion technology is used to extract the features of face images.Feature fusion technology based on PCA dimensionality reduction.The results show that the feature fusion technology after dimensionality reduction has achieved good results in both time and recognition rate.Subsequently,the influence of the DBN network on the recognition rate of complex illuminated face images was further studied.The experiment showed that the DBN network can greatly shorten the recognition time and improve the recognition rate after applying the LBP feature extraction technology.