|
Edge and detail of infrared image are blurry or loss after denoising with the threshold shrinkage arithmetic. A new
adaptive denoising and enhancing algorithm with detail enhancement based on a new Contourlet Transform with Sharp
Frequency Localization(CT-SFL) is proposed to preserve the edge better. CT-SFL has the characteristic of well-localized
in the frequency domain compared with the original contourlet. Firstly, CT-SFL, instead of the original contourlet, is
employed as the multiscale decompositon to decompose the infrared image into subbands. Secondly, the hierarchical
adaptive denoising threshold of new Contourlet coefficient is estimated respectively by each location from different scale
and directional subband, the noisy image is denoising with soft threshold related to the transform scale and direction,
then the denosing image is enhanced by taking decomposable scale and directional energy into account with
intrasubband and interscale dependencies. Thirdly, inverse CT-SFL is used to reconstruct the denoising and enhancing
image. Finally, in order to reduce significant amount of aliasing components which are located far away from the desired
support because of the new Contourlet Transform, cycle spinning is accomplished to the whole denoising and
enhancement process to overcome the lack of translation invariance property and suppress pseddo-Gibbs phenomena
around singularities of denoising image. Numerical experiments on infrared noisy image show that the proposed novel
algorithm can significantly outperform some arithmetics based on contourlet like 3 sigma, VisuShrink and Bayes
Shrinkage in all kinds of noise spectral density both in terms of PSNR(by several dB) and in visual quality, which can
enhance image's detail and stretch its contrast with nearly similar computational complexity.
|
