Abstract:To circumvent the visual distortions due to the discontinuity of the
curvelet hard-thresholding and the constant reconstruction deviation
resulted from the soft-thresholding, we considered the distribution
characteristics of noise coefficients in each curvelet subband and
the desired properties of an ideal curvelet thresholding scheme, and
developed a new thresholding function using Chi-square cumulative
distribution function. Further, in order to eliminate surrounding
effect inherent in curvelet thresholding denoising framework and
simultaneously achieve a better balance between detail preservation
and noise removal, a novel curvelet denoising method based on data
fusion theory is proposed. It is realized by operating the partial
differential equation denoising method on the small scales and our
thresholding method on the big ones respectively. Theoretical
analysis and simulation results show that the proposed denoising
method outperforms the soft and hard thresholding denoising methods
in terms of the denoising effect and visual quality.