Image spectrum decomposition and narrow-band noise removal based on 2D separable nonuniform filter banks

In this paper, a simple but effective method is proposed for image spectrum decomposition and denoising, due to the fact that the image cannot be decomposed flexibly in a transform domain. According to its spectrum distribution, the image decomposition is carried out by merging the corresponding filters of a uniform cosine-modulated filter bank, resulting 2D separable nonuniform filter banks. This method performs efficiently when the power spectrum of the image is distributed concentratedly. Moreover, if the image is corrupted by a narrow-band noise in the region where its power spectrum is low, the noise can be easily detected and removed by using the proposed idea of spectrum decomposition. A simulation of image Barbara is given to demonstrate that the proposed method is capable of achieving good performance in decomposing the power spectrum and removing the narrow-band noise involved.