Consider a grayscale image convolved with a bank of 2D wavelet quadrature pairs – in my case, log-Gabor filters. I have eight filters. For simplicity, let's say they are all vertically oriented, and an octave apart in frequency.

At some image location $(x, y)$ I now get 16 response values (one for each even and odd filter at eight different frequencies):

example plot of even and odd responses

I would now like to estimate the response at an intermediate frequency, say, at three-and-a-half octaves. Can that be done?

Intuitively, I would expect the local distribution of even and odd frequencies to be something to which I can (within obvious sampling limits) fit a function, allowing me to estimate intermediate responses. Is this correct? Can such an expression be derived from the filter kernel itself?


1 Answer 1


I have since found this paper, which provides a nice framework for how to think about this problem.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.