Font Size: a A A

A Study Of Nonlocal Total Variation For Image Denoising Based On SSIM

Posted on:2015-01-15Degree:MasterType:Thesis
Country:ChinaCandidate:T YuFull Text:PDF
GTID:2308330464466794Subject:Computational Mathematics
Abstract/Summary:PDF Full Text Request
In recent years, the math tools, especially represented by variational and partial differential equation, have become one of the basic tools for researching of image processing and computer vision. They are applied in all fields of image processing. This paper mainly deals with the establishment of mathematical model and the research on image denoising,by using these methods. These tools not only provide theory basis for modeling, but also make it convenient for us to do research on models’ performance and the effectiveness of algorithms. For denoising problem of image processing, structural similarity(SSIM) was introduced into nonlocal total variation(NLTV) image denoising model in this paper, which is inspired by the structural similarity index making an assessment of image quality effectively. Then we propose a new model based on SSIM and give a solution for it. Our main work is as follows:1.SSIM is introduced into the NLTV model, because when doing 2L measure of its loyalty,it doesn’t consider space structural of images. Then we propose a new image denoising model, which is called the first model, and give its corresponding solution, which is called the first algorithm. It is SSIM that is introduced into the loyalty instead of the 2L measure, which considers not only the pixel difference between images, but also the luminance and the contrast of the whole image. Thus the result of denoising conforms the visual effects of human eyes better. Compared with the NLTV model, which is computed by introducing the Split-Bregman, the first model not only contain more details in visual effects but also improve the effectiveness of denoising to a certain extent.2. The regularization in the first algorithm contain an adaptive weight function measured by MSE. When our processing detail-rich piece in the edge region, it is difficult for it to find pieces which are more similar to itself, resulting in no significant denoising effect. So it can not perform well in denoising the edge regions. To improve it, an adaptive weight function computational method was proposed, which is called the second algorithm. In the second algorithm, SSIM index was introduced to measure the similarity between pieces. Thus we can accurately estimate the similarity between pieces in edge regions and discriminate between the pieces with less similarities and the pieces without similarities, which improves the model algorithm’s function in the edge regions with complex structural information. Numerical experiments prove that compared with the first algorithm whose weight function is measured by MSE. That the second algorithm introduced an adaptive weight function can denoise effectively, simultaneously it keeps the structural information of the edge and improves the visual effect of the denoised images.
Keywords/Search Tags:image denoising, SSIM, NLTV, L~2 measure, Split-Bregman, adaptive weight function
PDF Full Text Request
Related items