Research On Fundus Image Restoration Based On Criminisi Algorithm

The fundus of the eye is the only place in the body where arteries,veins and capillaries can be visually and centrally observed with the naked eye,and these vessels can reflect the dynamics and health status of the entire blood circulation in the body.The fundus image can be used to clearly visualize the fundus.With fundus images,the tissue structure of the eye can be clearly observed and analyzed for abnormalities,and specific treatment plans can be given based on detailed screening and diagnosis.However,due to the design of fundus image acquisition instruments or interference from surrounding lights,the actual fundus images may show light spots,which may affect doctors’ clinical decisions.Therefore,based on image processing technology,it is of strong clinical application to study the algorithm for removing and restoring the light spots in fundus images.the Criminisi algorithm is a simple and efficient image restoration algorithm based on sample blocks.In this paper,we conduct research based on the Criminisi image restoration algorithm to explore new and improved image restoration algorithms,and the main research work is as follows.(1)In this paper,based on the Criminisi algorithm,the improved optimization algorithm is proposed by combining the luminance local absolute difference.Firstly,the luminance local absolute difference is proposed by introducing the luminance local variance,which can effectively distinguish the texture and flat area in the image,and introducing this algorithm into the priority calculation formula to improve the reliability of the priority calculation.Then the size of the sample block is selected adaptively based on the luminance local absolute value to achieve the restoration of different region structures of the image.Gradient similarity and luminance similarity are also introduced to improve the matching method of sample blocks to match the sample blocks that better match the visual perception of human eyes.The experiments show that the restoration effect of this improved algorithm has less restoration traces at the restored regions and better visual effects than other algorithms,and the overall transition of the image is more natural,and the objective evaluation values of PSNR and SSIM are higher.(2)The Criminisi image algorithm is improved based on the structure tensor with weighted filtering.Since the structure tensor contains information about the direction and magnitude of change in the image neighborhood,the eigenvalues of the structure tensor can determine different feature regions in the image,such as edge regions,texture regions and corner point regions.In this paper,we introduce the eigenvalues of the structure tensor into the data term and propose a new definition of the data term,which can better distinguish different structural regions of the image.Then the average correlation factor of pixel blocks is introduced and added to the matching algorithm of sample blocks to make the sample block matching algorithm more reasonable.Then the concept of weighted filtering is introduced to weight the final filled region by matching multiple candidate regions.Finally,according to the theoretical analysis and experimental steps described above,the experimental code is designed and experiments are conducted.The experiments show that the restoration effect of the improved algorithm has less restoration traces at the restored area and the overall transition of the image is more natural than that of the Criminisi algorithm and the algorithm proposed in Chapter 4 of this paper,and the objective evaluation values of PSNR and SSIM are higher.