imregister function and pyramid levels

Dear Alex

My question is related to imregister function which utilizes Gaussian pyramid levels. What I do is I decompose the image into its wavelets pyramids instead of Gaussian pyramids and then I register the wavelets coefficients which contains the detail and approximation detail images using imregister function.Now I set pyramidlevels =1 because I have already decomposed the images into wavelet pyramids . The mutual information value for optimization during registration of wavelets coefficients should be higher than that of guassian pyramids according to research papers but I am getting lower MI in case of wavelets based registration .Please let me know where I am making mistake or Is it the imregister() function which is causing the problem ?

Thanks

Hina

[Matt J]

The claims of the paper you mention seem doubtful to me. The difference in MI would depend on how much detail you've thrown away in each decomposition. You haven't mentioned how you've selected that, but it seems like it would depend greatly on the parameters of the pyramid and on the image data. Suppose, for example, that your reference and target images were composed purely of Gaussian lobes, so that the Gaussian decomposition sacrificed no detail. Then rationally, you would expect the Gaussian pyramid to have better MI than the wavelets.

[Matt J]

hINA Commented:

Thank you Matt for the quick response. I did a literature survey and found out that wavelet pyramids help in better registration particularly the LL component of one image once registered with LL component of the other image results in better optimization in terms of MI. One very significant research paper in this regard which compares the Gaussian pyramid with wavelet pyramids is mentioned below . Multimodal Brain Image Registration Based on Wavelet Transform Using SAD and MI Jue Wu1 and Albert C.S. Chung2

I took two images MRI and PET .Did simple image registration using imregister function with 4 pyramid levels and display the optimization parameters .The maximum MI was around 1.12

One the other hand I decomposed the same two images into their wavelet coefficients and registered two LL images (containing the maxi info)using im register function and kept the pyramid value =1 because I don't want the function to use pyramid now . I want to see the performance of wavelet pyramids only .The max MI value was around 0.67

Please let me know how do you see it ?