Abstract: A simple method for image spectrum segmentation and de-noising is proposed, due to the fact that the input image cannot be decomposed flexibly in a transform domain. Multi-channel linear-phase nonuniform filter banks with proper sampling factors are designed which are applied to rows and columns of the input image, respectively, according to its power spectrum distribution. By this procedure, the image can be segmented nonuniformly in the frequency domain. In addition, provided that the power spectrum of the noise involved concentrates on the low power part of the original image, it can be detected and removed easily by using the proposed idea of spectrum segmentation. This method performs efficiently when the power spectrum of the original image is distributed concentratedly. A design example is given to demonstrate the simplicity and effectiveness of the proposed method.

Key words: image spectrum segmentation, de-noising, spectrum distribution, nonuniform filter banks