Globalized BM3D using fast eigenvalue filtering

In this paper, we propose a progressive image denoising method using iterative filtering with Chebyshev polynomial approximation (CPA). It is known that a non-local/local image denoising method can be represented as matrix notation, and its denoising performance is improved by filtering the eigenvalues of the filter matrix. However, the eigenvalue filtering requires much computation time for eigendecomposition. To filter eigenvalues effectively, we proposed a fast eigenvalue filtering method using CPA [1]. The method drastically reduces the computation time but it still requires to construct a large sparse matrix. It often leads to much computational complexity. To overcome the problem, we propose an eigenvalue filtering method which does not construct a filter matrix by using the characteristic of the CPA. Experimental results show that our method is fast and applicable to large-size images. Additionally, the denoising performance of our method is almost better than those of the previous methods both in visual qualities and objective measures.