Denoising with Adaptive Eigenblock Dictionary Learning

Abstract

Denoising an image means to reconstruct a clean image (without noise) from a corrupted noisy one. The ability to efficiently and automatically identify which regions or pixels belong to the clean image(or denoised) is a powerful tool. Although many researchers have investigated ways to denoise an image, and have produced various denoising algorithms, challenges still remains. Many denoising algorithms loose some of the important local features of the image while getting rid of the noise. In this report we propose a new method for image denoising using Adaptive Eigenblock Dictionary Learning techniques. The algorithm captures the local spatial important features of the image by using neighbouring overlapping blocks for each block of a clean image to be reconstructed. The method takes advantage of the low rank dictionaries since a pixel belongs to many blocks. To reduce the size of local dictionaries and to speed up computations, the knowledge of principal component analysis is employed. The problem of denoising an image is transformed to solving a l1 minimization problem and the knowledge of sparsity are used for this purpose. The method shows a good performance in the sense that it denoises the images and doesn’t allow many features of the image to be lost. Performance of our algorithm is validated using a picture of Lena.

11 Figures and Tables

Cite this paper

@inproceedings{Nika2013DenoisingWA, title={Denoising with Adaptive Eigenblock Dictionary Learning}, author={Varvara Nika and Ying Wang and Prashant Athavale}, year={2013} }