Image Deblurring And Super-resolution Based On Image Prior Learning

As an important carrier of information in the era of big data,image data is one of the most important ways for people to understand the world.However,limited by the imaging equipment,imaging environment and other factors,the resulting image will lose a lot of details,unable to meet People's Daily life and production needs.Image deblurring and super-resolution are classical image post-processing technologies,which can improve the image clarity and resolution without hardware update.This paper is devoted to the research of image deblurring and image super resolution algorithm.The main research results are as follows:The first work in this paper is blind image deblurring based on convolutional spectral features and L₀-regularized,In order to improve the performance of traditional image deblurring algorithm based on L₀-regularized,a robust blind image deblurring algorithm based on convolution spectrum and L0 regularization prior is proposed.The algorithm adds the prior to the kernel,and ensures the good convergence of each sub-problem by separating variables and minimizing alternations.Experimental results show that this algorithm is better than the original one in the natural image deblurring task.The second work in this paper is deep image prior powered by regularization by denoising.In order to improve the performance of image super resolution algorithm based on DIP,an explicit denoising prior method is proposed to integrate the depth image prior into the DIP,so as to enrich the implicit prior and improve the effect of the original algorithm.Experimental results show that this algorithm is better than the original one in image super-resolution reconstruction...