Image Denoising Using Sparse Representation

Please enable javascript in your browser.

Beygiharchegani, Sajjad | 2010

632 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 40677 (05)
University: Sharif University of Technology
Department: Electrical Engineering
Advisor(s): Marvasti, Farokh
Abstract:
In this thesis, two novel image noise reduction approaches are proposed which can be implemented both in sparse signal processing domains such as learned dictionaries or wavelet and DCT. We first introduced a new probability density function (PDF) for the coefficients of image in transform domain and after that by using distinct thresholding function for each of coefficients we reduce noise in transform domain that is equivalent to reduce noise in time domain, since our transformation are unitary . In this scheme, we used variational approximation theory to find the optimum threshold values and noise variance simultaneously. In second method, we focus on impulsive noise reduction using sparse method to this end we assume that output image of a median filter is like an image which contaminated with Gaussian noise. Therefore it is possible to use iterative shrinkage Thresholding strategy to cancel noise from image. Experimental results indicate that both of proposed methods outperform several other established and state-of-the-are image denoising techniques, in terms of Peak- Signal-to-Noise-Ratio (PSNR) and visual quality in their class of noise.
Keywords:
Sparse Decomposition ; Noise Removing ; Gaussian Noise ; Impulsive Noise Removal ; Sparse Signal Representation