Hi,

could you explain me your idea why Contourlet transform is beneficial to other methods like normal wavelet transform or texture based analysis?

The Contourlet transform has its benefits more in describing pieces where you have contour like image content which is better represented than by normal wavelet transform. So do you think on representing the fine structure found within a Mammo-image and getting the microcalcifications by building the difference with the original image? 

Hi, do you want use a directionality image features ? contourlet decomposition can be very useful for example when you want to analyze directionality of textures structures into image. In the other hand, if you are interesting mostly to multiresolution analysis, then other wavelet transformation could give a better result.

yes lukasz i have planed to use directionality image features  in order to detect microclacification whether there is any article related to nonsubsampled Contourlet transform?

Thanks Peter, since microcalcifications are very tiny ,i want to extract the contour of each microcalcification so i have decided to use contourlet transform ,if there is any article related to non subsampled contourlet transform please send me.

Hi Vignesh,

I did years ago image compression, using the Contourlet transform. I think you already know the seminal paper of Minh Do "The Nonsubsampled Contourlet Transform: Theory, Design, and Applications". Not using decimation (i.e. nonsubsampling) has many benefits and prevents highly from artifacts. From my experience, the number of and calculation of the bandpass directional subbands crucial. Since structure within Mammo-images are of low contrast this is very important. I don't think that focusing on micro calcifications alone will work. There is the detection problem - yielding false positive and false negative.

Maybe, if you try to firstly estimate the background structure, you can then make a difference between estimate and (filtered) original in order to detect microcalcifications.

I did so with dental panoramic X-ray for denoising with preserving analytical content. I firstly estimated the varying noise, getting a "noise image" and then I subtracted the energies of the noise image wavelet coefficients from the original image wavelet coefficients energy using the wavelet space. Transforming the result back into spatial yielded the denoised image.

thank you Peter