A Gabor based technique for image denoising

As an alternative to the wavelet, Gabor function has been used as an efficient representation of two dimensional signals. We are interested in BayesShrink techniques for image denoising, and have shown in our previous work that BayesShrink Ridgelet performs better than VisuShrink ridgelet and VisuShrink wavelet. In this paper, a dyadic Gabor filter bank is combined with BayesShrink method for image denoising. In the proposed method, the noisy image is decomposed to different channels in several levels by a dyadic Gabor filter bank. To recover the image, the corrupting noise is removed by applying the proposed BayesShrink method on the noisy Gabor coefficients. The noise variance is estimated in Gabor domain and the estimated noise is then used to dynamically calculate an individual threshold for each spatio-frequency channel. Finally denoised coefficients are transformed back to reconstruct the image