Noise reduction of images with multiple subband transforms

It is reported that the use of multiple subband transforms for thresholding based denoising gains performance in the mean square error sense. In traditional thresholding-based methods, a noisy image is decomposed by linear transformations such as wavelets, FFT and so on and the transformed coefficients are hard/soft-thresholded. In particular, it is well-known that wavelets work well for denoising. From the viewpoint that wavelets are in a class of subband transforms, we propose a strategy in which multiple subband transforms are switched region by region, i.e. block by block. For reconstruction, the projection-based iterative method is used. Experimental results are pretty good and promising.