How to validate high pass filtered images with original image?

I want to compare my original image with the high pass filtered image.Here I want to validate the high pass filtered image using a measurement.Can you suggest any specific method to validate it?

As an example how to identify illumination removal of an image after high pass filtering?

Answer 1

[Bjorn Gustavsson]

Well "validate" a high-pass filtered image to the original image is a slightly confusing concept to me - by high-pass filtering you make a signifficant change to your image. Why not start with something simple:

subplot(2,2,1)
imagesc(Orig_IM)
subplot(2,2,2)
imagesc(hpfiltered_IM)
subplot(2,2,3)
imagesc(Orig_IM-hpfiltered_IM)
subplot(2,2,2)
imagesc(log(abs(fftshift(fft2(Orig_IM-hpfiltered_IM)))))

Then you can "validate" visually - perhaps good enough, ought to be for image processing objectives, you do this type of filtering "for the eye"...

HTH

[Image Analyst]

I second this. Validate is not defined.

If you remove the illumination, the scene will be black.

Perhaps you mean to reduce the illumination non-uniformities. However a high pass filter will not only eliminate slowly varying intensity changes in the illumination, but also the slowly varying changes in the subject/target/object in the scene that you're interested in (if any are there that you care about).

[Nayana R]

Yes I mean the difference between original image and the image after reducing illumination.How we can show the difference between these two images?Can i use a histogram or entropy value to find the difference in intensities of two images?

[Bjorn Gustavsson]

Sure, knock yourself out! Entropy is defined as , and I fail to see how that could give you all that much interesting information here. Perhaps you get something out of it perhaps you get something better if you normalize the images with their mean intensities, perhaps you get something more interesting if you calculate the entropies of del2(I/mean(I(:))), but I fail to see what fun could come out of that. Maybe you get something out of looking at the histograms of del2(I/mean(I(:))). Maybe you get something interesting by looking at the difference between the original and filterd images after passing them through histogram equalization.

But the only thing you do with this type of high-pass/homomorphic filtering is to reduce the low-frequency components in the image, that's it, so I expect the only result to come out is something like "yeah, the homomorphically filtered image has more pronounced small-scale features".

Finally: have you even tried my suggestion?

HTH

[Nayana R]

Yes I tried your suggestions.Thankyou for that.but i cannot understand HTH?

[Nayana R]

For the project that i am doing now our supervisor asked to validate the high pass filtering of my images.That's why I want to find the difference between these images.

[Bjorn Gustavsson]

Well, then go ask your supervisor what he means with validate - you clearly don't know, we others guess around. I've coughed up a couple of suggestions. The thing here is that, and this is explicit and concious repetition, homomorphic high-pass filtering is something you do "for the eye", and it is best to the "validation" (whatever that means) by displaying the images, their differences, and a couple of different transformed images (fft-spectrum, histogram equalized...) and see what happens.

HTH - hope this helps

[Nayana R]

Thank u for your suggestions Bjorn Gustavsson.