Research On Enhancement Method Of Face Recognition At A Long Distance

In recent years,the extensive and in-depth application of deep learning theory in the field of computer vision has greatly promoted the development of image processing and face recognition technology.However,although the development of standard face recognition technology with the cooperation of users has matured,the long-distance face images in practical applications have low quality,insufficient resolution,and low brightness due to imaging equipment,lighting,and other factors.Such problems have greatly affected the accuracy of face recognition algorithms.Therefore,it is necessary to enhance the long-distance face image to improve the recognition rate.This paper studies face recognition technology and image enhancement technology based on the combination of traditional algorithms and deep learning theory.The main work is divided into the following three parts:First of all,based on the investigation of long-distance face image recognition technology at home and abroad,this article has conducted in-depth research on the basic theory of image enhancement theory and deep learning.In terms of image enhancement theory,the traditional enhancement algorithm theory of spatial and transform domain enhancement is introduced.In terms of deep learning,this article delves into the basic structure of convolutional neural networks: convolutional layers,pooling layers,fully connected layers,and commonly used activation functions,which lays a foundation for the enhanced algorithm scheme that combines traditional algorithms and deep learning theory basis.Secondly,based on the different problems faced by long-distance face images,this paper uses the existing enhancement algorithms to solve them one by one,and studied a simple parameter G-Log enhancement algorithm,which uses simple mathematical transformation to achieve image brightness and contrast Enhancement.And after summarizing the advantages and disadvantages of each algorithm,the Retinex-CNN enhancement algorithm is studied.This algorithm uses convolutional neural network to learn features and uses Retinex theory as the theoretical guide for the construction of the loss function.The algorithm uses VGG network to learn features and uses Retinex theory as the theoretical guide for the construction of the loss function.Combined with BM3 D and fractional order image enhancement theory,it has achieved good enhancement effects.Finally,this paper uses VGG-Face to extract features from long-distance face images,and then uses the SVM algorithm to classify the extracted features to calculate the recognition rate.In a large number of experiments,because deep network training requires a large amount of long-distance face image data,and the proposed Retinex-CNN requires paired normal / low-light images as input training,in this paper,through a variety of data expansion methods and statistically generating simulated low-light images from the pixel distribution of face images in low-light situations,the database of existing longdistance face images is expanded and reconstructed.In summary,this paper studied the Retinex-CNN enhancement algorithm on the basis of combining multiple enhancement algorithms.And this paper combined with the existing small sample database structure to build a long-distance face image database through data expansion to facilitate the training of the long-distance face image enhancement recognition system.Many experiments show that the enhanced recognition method proposed in this paper greatly improves the recognition rate of long-distance face images.