Medical image denoising using sparse representations
Journal
Proceedings - 2017 IEEE 8th International Conference on Awareness Science and Technology, iCAST 2017
Journal Volume
2018-January
ISBN
9781538629659
Date Issued
2017-07-01
Author(s)
Abstract
Image noise reduction, or denoising, became an attractive research domain in the recent years. There are several published approaches in the image denoising field and each one has its presumptions, advantages, and limitations. In this paper, we address the problem of eliminating or even decreasing the noise from medical images. Some schemes for image noise reduction are proposed. The schemes include DCT OMP, DCT BOMP, Log Gabor BOMP, DCT OCMP and Wavelet OMP. In these schemes, the image content is represented as a sparse linear combination of a set of atoms that can be obtained from some trained dictionaries. In literature, several algorithms are proposed to build up these kinds of dictionaries. In our study, the K-SVD algorithm is utilized to obtain a dictionary that can effectively describe the image. Also, some greedy algorithms are used to perform the sparse-coding of the signal. The proposed schemes have been tested on two different medical images with different levels of an additive noise. The Log Gabor BOMP scheme showed a significant outperformance, in both the de-noising results and computational time, on all the other schemes.
Subjects
dictionary learning | K-SVD | Medical images denoising | sparse representation
SDGs
Type
conference paper