Image noise reduction by dynamic thresholding of correlated wavelet intensity and anisotropy

Abstract

A novel image noise reduction filter is developed in wavelet domain. The correlations of intensity and anisotropy of wavelet coefficients in different sub-bands and scales are utilized as features to separate signal from random noise. Dynamic thresholding is used to further increase the sensitivity and discrimination against different noise patterns and standard deviations. Simulation in Matlab is carried out by filtering an AWGN-added sub-set of test images from Kodak image data base and a standard digital Macbeth Color Chart. Average PSNR scores of R, G and B color channels are competitive to bilateral filter and a prominent commercial noise reduction tool Neat Image. In addition, the low computational complexity of threshold operation makes it applicable for low cost implementations in various imaging devices as a real time noise reduction module.