Diffusion-based image denoising combining curvelet and wavelet

In this paper, a new denoising technique for images corrupted with additive white Gaussian noise is presented. The technique used here is to combine the wavelet transform and the curvelet transform with anisotropic diffusion. Curvelet transform is a new geometric-based multiscale transform developed to give sparse representation of images with singularities along curves. The wavelet transform is known to be good at processing point singularities and small patches. Thus a combination of these two transforms can be used for image denoising such that the result image would contain the information considered as significant by any of the two transforms and thus obtaining a visually improved denoised image. One of the problems faced in image denoising techniques is the presence of pseudo-Gibbs artifacts. To reduce these artifacts the denoised image can be further processed by anisotropic or nonlinear diffusion, in which only the high-frequency part of the image is changed. The experimental results obtained from various images show that this method gives better performance when compared to curvelet-based denoising and wavelet-based denoising.