Discrete sine transform shrinkage functions based image denoising
Kai Xie · 2011
A novel image denoising technique is proposed with shrinkage functions learning in discrete sine transform (DST) domain. The technique uses the regularized least square method to compute optimally the transform coefficients of DST in patches of example images. Once the shrinkage functions have been gotten by train, they can be used directly to new images that are suffering from an additive noise with the same power as the learned image. The method has no to know the prior mode of the noisy image beforehand. If these images are similar to the ones the functions were trained on, the performance of the overall denoising is expected to be very good.