Research On The Application Of Sparse Representation And Non-local Prior In Image Processing

It is inevitable that the image will be affected by various external or internal factors during the acquisition or transmission process,resulting in the degradation and quality of the image.In order to restore the original image for subsequent image processing and application,many scholars have done a lot of research work,of which sparse representation and variational models are relatively common and classic algorithms among a large number of image algorithms.In this paper,according to the degradation model of the image in different situations,a sparse representation color image denoising algorithm based on intrinsic image decomposition and a weighted variational image defogging algorithm based on non-local priors are respectively proposed.The main research contents and innovations are as follows:(1)The research background and significance of image denoising and dehazing algorithms and the current status of research are briefly introduced.Combining with the previous research results,the degradation model of the image when it is polluted by noise and the hazy weather degradation model of the image in the case of haze are respectively explained.From these two degradation models,the research focus of image denoising and dehazing algorithms can be clearly understood,so this article briefly introduces some representative image denoising and dehazing algorithms.In addition,in order to verify the denoising and dehazing capabilities between different algorithms,this article roughly introduces the evaluation indicators of image quality.(2)Due to the different characteristics of the reflectance map and light map of a color noisy image decomposed by the intrinsic image algorithm,this paper uses K-SVD color image denoising algorithm and non-local sparse representation gray image denoising algorithm process these two parts separately.Then,the processed images is synthesized by an intrinsic image decomposition algorithm to obtain a denoised result image.The experimental results show that the sparse representation color image denoising algorithm based on intrinsic image decomposition is more effective than some classic image algorithms or the gray image denoising algorithm applied directly in the color channel.It has the ability to retain more image details while improving the image denoising effect.(3)According to the hazy degradation model of the image,this paper first calculates its atmospheric light estimate for a hazy image and uses non-local prior knowledge to estimate the initial transmittance value of the image,and then establishes a weighted variational model to optimize the transmittance estimate value to obtain the optimized transmittance estimate.Finally,the atmospheric light estimate and the optimized transmittance estimate are applied to the hazy degradation model to recover the haze-free image.Through experiments,it is found that the weighted variational image dehazing algorithm based on non-local priors can obtain clearer haze-free images with better universality and retain more image details.