Research On Enhancement Algorithm For Low Quality Image

With the popularity of modern network technology and imaging devices,image as one of the main ways of communication for people,plays an increasingly important role in life.Due to the limitation of shooting environment and imaging devices,the captured images often appear poor problems,such as brightness reduction,color degradation,blurring and low identification.In order to extract the key information and improve the usability of the images,it is urgent to enhance the low-quality images.This paper mainly focuses on the research of backlight image enhancement and motion blur image restoration.The main contributions are as follows:(1)Backlight shooting often causes the problems of brightness reduction and loss of detail information.Thus,this thesis proposes an image enhancement algorithm by fusing global and local brightness enhancements,for the above issues.Through the color estimation model(CEM)based on global statistics,the detail information of backlight image can be well restored.To avoid the fuzzy effect on dark region of backlight image by CEM,the brightness preserving color estimation model(BPCEM)is presented.By taking advantages of both CEM and BPCEM model,an adaptive fusion method based on block information entropy is proposed.(2)In this thesis,a license plate image deblurring model based on adaptive blur kernel is proposed,to solve the blur problem caused by the relative motion between vehicle and camera.To restore the license plate images of different blurriness,an adaptive blur kernel detection model based on multi-scale blur kernel space is presented.Then,to improve the clarity and identification of license plate image,a regularization method based on statistical features of gradient and intensity is proposed.(3)The classic image restoration model based on the hypothesis that noise follows Gaussian distribution is sensitive to different types of noise.Thus,the regularization method which introduces the Laplacian constraint is proposed.Besides,a blur kernel estimation model based on salient edges of structure component is presented,to weaken the adverse influence caused by image redundant edge during kernel estimation.Finally,experiments on the synthetic database and public database of low-quality,shows the good performance of the proposed method.