Wavelet denoising in non gaussian noise using MDL principle

Wavelet methods have succeeded in image denoising in gaussian noise. In non-gaussian noise, however, these methods will degrade drastically. By employing the MDL principle and a wavelet coefficient model, this paper discusses a study on image denoising in gaussian mixture noise. A new denoising scheme is derived and is based on per pixel detection. Experiment results show that the new scheme can not only denoise the image with small square error, but also provide a facility for the further compression.