Research On Anisotropic Diffusion Image Enhancement Based On Geometric Features

Image enhancement is one of the basic contents of image processing.Many image processing problems have directional flow structures.Common image enhancement methods have the disadvantages of stepping,mottled or false edges when dealing with such problems,and can blur important features such as edges.In this case,it is desirable to have a tool that can improve the quality of the flow-like structure without destroying important detail features.This paper mainly studies the image enhancement problem by constructing an anisotropic diffusion model based on geometric features.The goal is to complete the interruption line or enhance the flow structure.A reliable tool for analyzing coherent flow structures in the past is the gradient?_,which is not suitable for processing parallel structures,although it is useful for detecting edges.The basic idea of the geometric framework is to embed the image as a high-dimensional surface,and to view the image as a three-dimensional surface.From a geometrical point of view,the image enhancement problem is equivalent to finding an approximate segmented smooth surface.The second basic form describes the shape of the surface—the extent to which the surface deviates from the section,and is a good way to characterize the geometry of the surface.We use the second basic type quadratic matrixinstead of the?_structural tensor_to extract the non-local structural features of the image itself,and propose an anisotropic diffusion image enhancement model based on the second basic type.The eigenvalues of the model structure tensor contain rich structural information.Based on this,the metric of local coherence of the image is proposed for image partitioning,and the image is divided into smooth regions,flow-like regions and corner/T-shaped regions.In order to enhance the coherence of the flow-like region,and use the structural features of the image to complete the enhanced diffusion process,we adjust the structural tensor_to the diffusion tensor,and propose the diffusion mechanism of the control model.The eigenvector ofcontrols the direction of the model's diffusion,and the eigenvalue determines the smoothness with the image coherence during the smoothing process.In addition,two different diffusion schemes,adaptive diffusion scheme and filtered diffusion scheme,are designed to guide the diffusion process of the model.Compared with the adaptive diffusion scheme,the filtering and diffusion scheme considers the image information refinement partition more comprehensively,and can better guide the diffusion process according to the image information characteristics.In the numerical dispersion and experimental part of the model,the discrete formats of the finite difference method and the additive operator splitting(AOS)method are given respectively,and different kinds of test images are selected and tested by MATLAB.The experimental results show that the proposed model can achieve the effect of enhancing the quality of the flow-like structure,and protect the important information of the image,prevent the image distortion from being enhanced,and will not destroy the details of the original image.