Non-uniform Illumination Face Image Enhancement Based On Deep Convolution Neural Network And Recognition

Non-uniform illumination as one of the interferences in face recognition severely restricts the improvement of face recognition rate.Although the current face recognition algorithm can achieve 100% recognition rate under ideal illumination,the illumination under unconstrained conditions seriously affects the performance of face recognition algorithms,and this in turn affects the promotion and application of face recognition technology.In areas such as smart monitoring,in addition to requiring the system to have a high accuracy,it is also necessary to obtain visually clearly identifiable face images to facilitate manual verification at a later stage.However,illumination changes,especially extreme non-uniform illumination changes not only reduce the accuracy of face recognition algorithm,but also lead to people unable to obtain a clear face image,which brings great difficulty to the later manual verification.In recent years,with the development of deep learning,it has been widely used in the field of image noise reduction and enhancement.In order to obtain uniform illumination face image from nonuniform illumination one and improve the accuracy of face recognition algorithms,this paper has proposed two non-uniform illumination face image enhancement algorithms based on deep learning methods and has realized face recognition by extracting different illumination robust features of the enhanced image.This paper focuses on the research of non-uniform illumination face image enhancement algorithm based on deep convolution neural network,and extracts three kinds of illumination robust features for face recognition,aims to obtain uniform illumination face images and improve the accuracy of the recognition algorithm.The main work of this paper can be described as follows:(1)Based on multi-scale Retinex theory,a convolution neural network for nonuniform illumination face image enhancement is designed.The multi-scale Retinex algorithm is a classical algorithm for enhancing nonuniform illumination images.It can effectively improve the illumination distribution in image,but it often leaves boundary contours at the junction of light and dark areas of image,and it also has problems such as insufficient enhancement of areas with rich details.However,the visual imaging model embodied in the multi-scale Retinex algorithm has a perfect interpretation of the impact of illumination on face images.Inspired by the visual imaging model and the image enhancement process in the multi-scale Retinex theory,a non-uniform illumination face image enhancement convolution neural network is designed by this paper to restore the non-uniform illumination face image to uniform illumination one.In order to suppress the contours of the bright and dark regions in nonuniform illumination face images while retaining more face contour information,this paper obtains gradients in multiple directions for enhanced images and label images,and designs multi-directional gradient loss functions.The experimental results prove the effectiveness of this algorithm.(2)A Generative Adversarial Network for non-uniform illumination face image enhancement is designed.Since its emergence,the Generative Adversarial Network has been widely used in tasks such as style migration,image generation,and super resolution enhancement,and has achieved pretty good results in these fields.This paper proposes for the first time the use of a Generative Adversarial Network for non-uniform illumination of face image enhancement.This section designs a Generative Adversarial Network with a non-uniform illumination face image enhancement network in(1)as a generator,and introduces perceptual loss to construct the loss function of the generator at the feature level in order to further enhance the quality and achieve the overall optimization of enhanced images.(3)Extracting different illumination robust features from the enhanced face images for face recognition.In order to verify the performance of the two enhancement algorithms,this paper compares the recognition rate of the same recognition algorithm on the original images and the enhanced images.In addition,the performance of the face recognition method combined with the enhancement algorithm and the recognition algorithm and other face recognition algorithms for non-uniform illumination is also compared.It is confirmed that the enhancement algorithm can make the existing face recognition algorithm get a higher accuracy.