Total Variation Based Wavelet Domain Filter for Image Denoising

Abstract

In this paper the method total variation (TV) is applied on noisy image decomposed in wavelet domain for removal of additive white gaussian noise (AWGN). LL subband of a single decomposed noisy image is used to find the horizontal, vertical and diagonal edges. Using the pixel position of horizontal edges, the corresponding wavelet coefficients in HL subband is retained thresholding others to zero. Adopting the same procedure the vertical and diagonal details of LH and HH subband is retained. The method TV is applied to LL subband for one iteration only. Applying inverse wavelet transform on modified wavelet coefficients we get back the image with little noise. This little noise can be removed using TV filter with single iteration. The method performs well in terms of peak signal to noise ratio (PSNR) over many well known spatial and wavelet domain methods. The method also retains the edges and other detailed information very well.