Research On Low Illumination Face Recognition Algorithm Based On Retinex And Feature Fusion

Face recognition has been widely applied in the verification of identity information.This method has great research value due to its simple collection,fast recognition speed,and high security.However,in low illumination indoor environments,night time,and rainy outdoor environments,face images still have serious edge halo artifacts,inability to extract facial image features,and low accuracy in identifying human identity information after processing with traditional illumination enhancement algorithms.To address the above issues,the main research content of this article is as follows:(1)An improved Retinex facial image enhancement algorithm is proposed to address the increased illumination of low illumination facial images.First,the Bilateral filter algorithm is proposed to replace the traditional Gaussian filter to denoise the image and maintain the edge information of the face image;Secondly,in response to the severe situation of image halo artifacts,a transformation method based on local standard deviation was studied to achieve the elimination of halo artifacts;Then perform tone reconstruction on the image;Finally,this article constructed a low illumination face dataset based on the Extend Yale B and CAS-PEAL face datasets,and simulated and compared the feasibility of the improved Retinex face image enhancement algorithm from both subjective and objective aspects using MATLAB 2014 b.(2)After enhancing the illumination of low illumination facial images,there is still a problem of difficulty in feature extraction.Therefore,this paper proposes a feature extraction algorithm based on Gabor and improved multi threshold LBP fusion.Firstly,the scale and direction of the Gabor wavelet algorithm are selected to extract more local facial features;Secondly,in the traditional multi threshold LBP algorithm,3 × Based on the range of 3 neighborhoods and threshold of 0,it is proposed to use 7 × 7 neighborhood range and adding threshold T to improve the ability to extract texture details;Then,the focus was on the selection of eigenvalues and feature dimensions in the co-occurrence matrix;Next,research the blocking problem of facial images under different datasets;Finally,the improved Gabor filtering algorithm is fused with the improved multi threshold LBP algorithm,and simulation analysis is set in the constructed low illumination face image dataset to compare the advantages of the improved fusion algorithm over LBP algorithm,Gabor algorithm,and multi threshold LBP algorithm in feature histogram extraction,image information entropy value,and face recognition rate.Solved the problem of traditional feature extraction algorithms requiring multiple low illumination images to recognize facial images with low recognition rate.(3)Comparative simulation experiments were conducted on the recognition rate and recognition speed of the improved low illumination face recognition algorithm proposed in this article and the classic face recognition algorithm in the Extend Yale B and CASPEAL face image datasets,respectively.The feasibility and rationality of the improved algorithm in this article were verified through comparative simulation analysis results.Figure [60] Table [14] Reference [83]...