Restoration And Recognition Of Degraded Images Based On Prior Information

With the development of science and technology,image processing technology plays an increasingly important role in various fields.High quality,clarity,and accuracy of images are essential for many applications.However,due to the unique nature of image data and the massive amount of data,images are often subject to various interferences.Therefore,researching how to restore,detect,and recognize distorted images is a hot topic in the field of image processing.Prior information can be knowledge about the structure,texture,color,and other aspects of an image.It can be based on data statistics or physical knowledge,or it can be information obtained from processing the image,such as denoising,deblurring,and low-light enhancement.This thesis utilizes prior information to investigate the problems of degraded image restoration and recognition.The main research conducted in this thesis is as follows:(1)A method for degraded image restoration and recognition using sparse and low-rank priors is proposed.By combining sparse representation with low-rank representation priors,effective restoration and recognition of noisy images are achieved.This method fully utilizes the advantages of sparse and low-rank representations,providing higher computational efficiency and restoration accuracy when dealing with degraded images.Experimental results demonstrate the good performance of this method in various degraded image restoration tasks,providing strong support for further enhancing the application value of image processing technology.(2)By combining the prior knowledge generated from the physical model of orthogonal decomposition with MEF-Net,higher-quality low-light enhanced images are obtained.In practical applications,images captured in low-light environments often suffer from uneven illumination.To address this issue,the method of generating multiple exposed images through orthogonal decomposition is used,and MEF-Net is employed to fuse these images to achieve low-light enhancement.Experimental results show that compared to traditional low-light enhancement methods,this approach better preserves image details and texture information,providing more reliable data support for subsequent image analysis and detection tasks.(3)The proposed low-light enhancement method is integrated with the Pyramid Box face detection network to form an end-to-end network,enabling more accurate low-light face detection.Face detection has wide applications in computer vision,but in low-light environments,illumination can vary significantly across different regions,which often affects detection performance.To improve the accuracy of low-light face detection,the previously proposed low-light enhancement method is applied to the Pyramid Box face detection network.Comparative experiments demonstrate that this method achieves higher face detection accuracy in low-light environments.