The Research Of Enhancement Algorithm For Over Exposure And Low Illumination Images

Visual information which mainly composed of images is indispensable to the existence and development of human.Now the importance of image processing technologies in people's life is increasing daily,such as video monitoring,medical imaging,fingerprint unlock,production automation,military track and so on.Images with good illumination,fine texture and color ensure the normal operation of modern artificial intelligence technologies.However,we often get over-exposure or low illumination images due to the limitation of equipment or background conditions,these images seriously affect the intelligent technologies.This thesis mainly focus on the enhancement of colorful over-exposure and low illumination images,which supported financially by the National Nature Science Foundation of China(No.51405320).Over-exposure images usually lose the texture and color easily because of high illumination,and low illumination images often have fuzzy details and edge.Existing enhancement algorithms can enhance the contrast of these images,however halo,fading and over-enhancement phenomena usually happen during image processing.In this thesis,we solve the problem of texture and color loss in the over-exposure images,fuzzy details and low-contrast in the low illumination images by the proposed or modified algorithms.Firstly,this thesis introduced some existing image enhancement algorithms and frequently-used image quality assessment algorithms.Classic image enhancement algorithms include linear and nonlinear image enhancement algorithm,histogram correction,image denoising enhancement algorithm and Retinex algorithms.In order to select the appropriate enhancement algorithms,the problems of images and the characteristics of algorithms need to be taken into consideration.Image quality assessment algorithms include the mean value,standard deviation value,entropy value and peak signal to noise ratio,BRISQE value and NIQE value.Secondly,this thesis studied the texture and color restoration algorithms of local over-exposure images.HSV(Hue,Saturation,Value)was used to decompose the images as HSV can express the light,hue and bright degree of the images intuitively.We detected the over-exposure region with the modified saturation threshold value.Then region-filling algorithm was used to restore textures,and the priority of boundary point was weighted calculated by combining the light mean square value,data item and confidence to determine the repair order.Finally,the hue and saturation in over-exposure region were recovered by use the weight of space distance and luminance difference.Experimental results showed that the texture and color were recovered in local over-exposure region and the histogram distribution characteristic was better.Thirdly,this thesis studied the low illumination image detail enhancement algorithms.The low illumination image details are not distinct due to the low brightness.The low illumination image was decomposed into structure layer and detail layer by establishing gradient sparse and least square constraint model.Then the details of the image was strengthened by proposed multi-scale edge protection detail enhancement algorithm and filtered.The experimental results indicated that the low illumination image details were enhanced effectively.Finally,this thesis studied the low illumination image illumination enhancement algorithms.The algorithm is composed of two parts: one was the image decomposition in HSV color space,firstly V channel was decomposed by Wavelet Transform,then detail clear and high resolution gray image was obtained after high and low frequency coefficient being processed respectively;The other was the contrast improvement of V channel by using the Retinex algorithm based on guided filter.Finally,integrated V channel with other channels and output the repaired images.Experimental results of the image enhancement algorithms proposed in this thesis showed that the texture and color were restored in the over-exposure images,and the details were enhanced and the contrast was improved in the low illumination images.The visual information of these images was improved greatly,which was of great significance for image processing.