Variable block wavelet wiener estimator in image denoising

Abstract

This paper improves the Wiener filter in the wavelet domain without the usual zero mean assumption. An improved LESE (local expected square error) formula is derived. For each block size, the LESE and LESEimproved are computed. The block size that gave the minimum LESE is chosen. For that block if LESE is less than LESEimproved then the zero-mean filter is used else the non-zero mean filter is used. The improved filter gave a higher PSNR for all the test images and all the noise variances that we have used.