Design Of Checkerboard-shaped Filter Banks And Their Applications In Image Denoising

Extracting the multi-directional information of images has lately attracted much attention in image processing. At present, the wavelet transform and the directional filter banks are in common use. However, the wavelet transform can extract only three directions, and most subbands of the cosine filer banks can not extract directional information. Firstly, we design two 2-D checkerboard-shaped filter pairs with direction of 45 and 135, then construct two kind of directional filter banks via the wavelet transform and the cosine modulated filter banks. Six directional features of the image are extracted simply and effectively by cascading the checkerboard-shaped filter pairs and wavelets. The method is implemented in the wavelet-based image denoising by thresholding and Wiener filtering with elliptic directional windows. The experimental results show that the proposed algorithm can achieve significant improvement than the existing image denoising algorithms using the 2-D separable wavelets. We construct a new directional filter bank with flexible direction and frequency selectivity by cascading checkerboard-shaped filter pairs and the cosine modulated filter bank. First, a checkerboard-shaped filter pair is used to decompose an input image into two images contained different directional information of the original image. Next, a 2-D separable cosine modulated filter bank is applied to each of the two images for directional decomposition. This new structure equals to a new directional filter banks with redundancy ratio of 2. As its application, the BLS-GSM algorithm for image denoising is extended to use the new directional filter banks. Experimental results show that the proposed directional filter banks achieve better denoising performance than the methods using other directional filter banks for images of abundant textures.