A modified curvelet-like transform with application to image denoising

According to the essential ideas behind the curvelet transform, we recognize that the noise averaging property of the projection-slice theorem plays an important role in curvelet denoising as well as ridgelet coefficients thresholding. Based on this idea, a simplified curvelet-like transform is presented in this paper. Instead of using the original block size 2^-j2/ for subband [2^J, 2^f+1], it takes the full size of the image as the block side-length in every subband. Its denoising threshold for ridgelet coefficients of each subband is also adapted to the corresponding scale to avoid gray level increase and detail degradation. As a result, this modified implementation provides high speed processing. Furthermore, experiments with noisy images and poor quality X-ray pictures demonstrate that the suggested method works well when the noise deviation is estimated correctly.