Font Size: a A A

Anisotropic Diffusion Filtering Algorithm Based On Image Features

Posted on:2015-03-07Degree:MasterType:Thesis
Country:ChinaCandidate:D LiuFull Text:PDF
GTID:2268330428476014Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
Image filtering is an essential part of image preprocessing. Traditional filtering algorithms use template for image filtering, these algorithms blur the edge of image when denoising. Anisotropic diffusion filtering algorithm treats gradient as edge detection operator, and implements different degrees of diffusion for edge and non-edge regions, it protect the edge of image when denoising. Comparing with traditional filtering algorithms, anisotropic diffusion filtering algorithm has a great advantage in terms of edge retention.A large number of improved models are proposed by scholars to solve the problem which caused by anisotropic diffusion model, but most models are designed for diffusion equation instead of image feature information. This paper mainly focus on how to extract useful feature information, improving model based on the extracted feature to get better diffusion effect. The main contributions of this paper are as follows:Firstly, proposed an anisotropic diffusion equation based on the direction features of image. We take singular value decomposition for gradient matrix of image to get the direction feature. Based on the direction feature, we design weighting function to extend the diffusion to eight directions and ensure that each pixel implements adaptive diffusion on eight directions. New algorithm can control the diffusion direction, so it can protect the edge information of image better when denoising.Secondly, proposed an anisotropic diffusion equation based on the phase coherence feature. Phase coherence can accurately describe the characteristics of the image edge information, this paper uses log-Gabor wavelet to extract the phase coherence feature of image. Based on phase coherence feature, we design weighting function to ensure that different pixels achieve different degrees of adaptive diffusion. This new algorithm can control diffusion speed better between edge point and non-edge point, so it can protect the edge information of image better when denoising.
Keywords/Search Tags:anisotropic diffusion, directional characteristics, singular value decomposition, phase characteristics, log-Gabor wavelet transform, weighting function
PDF Full Text Request
Related items