Research on Random Noise Attenuation in Transform Domain

Random noise attenuation method by contourlet transform is researched in this paper. Compared with wavelet transform, the contourlet transform proposed by Do and Vetterli offers a high degree of directionality and anisotropy, and it represents images sparser. Consequently, contourlet transform can obtain better denoising result than wavelet transform when a simple thresholding scheme is applied on the contourlet coefficients. But Contourlet-based image denoising method always introduces artifacts due to the lack of translation invariance of the contourlet transform. And hence, cycle spinning is often used to compensate for the lack of translation invariance of the contourlet transform. Experiments on test images show that the contourlet transform denoising method using hard threshold can obtain about 1~2 dB higher PSNR values than classical 2-D separate wavelet transform. And denoising progress by contourlet transform with cycle spinning technology can reach more outstanding denoising results than that of without.