Bounds of a band pass filter for image processing

[jsr9119]

Hi all,

I'm working on some image analysis as a part of my research, specifically trying to match images. The method I am using transforms the image into the frequency domain and then applies a band pass (or mesa) filter to eliminate noise and the dc component of the image.

I have never done this kind of work before so maybe I am on the wrong track, but for the filter, I don't understand how the bounds of the filter should be assessed, i.e where the cutoff frequencies should be or how fast the cutoff should grow. I've basically been running for-loops to check all the different values, but my gut is telling me there should be a way to calculate these values. Am I wrong? Any insight is greatly appreciated.

Thank you,

Jeff

 

[jsgruszynski]

There frequency for an image filter refers to the frequency of change of pixels as a function of x and y. Something that is minimum frequency (~0) will be when a pixel level is constant across the entire axis. The maximum frequency (~1/N) will be when pixels alternate.

If your noise is pixel-by-pixel (and if your image doesn't change that fast, then that would be where you've put your cut-off frequency (between the two). Generally you'll probably use a low-pass filter rather than a band-pass filter (most noise is high frequency while most signal is low frequency - though there is no hard-and-fast rule about that).

 

[jsr9119]

That's very helpful, thank you. Do you know of any good books or articles that would go further in depth?