# Microcontrast

Started 4 months ago | Discussions thread
Re: Microcontrast

AiryDiscus wrote:

Joofa wrote:

Doctor J wrote:

J A C S wrote:

Actually, at one point only - (0,0) where you get the sum of all pixels. A far better representation can be obtained by subtracting the mean before taking the FT. One can still use a log scale but a much weaker one and get a better idea of the frequency distribution.

Another "optical illusion" comes from the fact that the frequencies are inversely proportional to the detail. A huge part of the FFT plot (aside from a centered square with 1/2 the side) represents a few pixels detail, roughly speaking. Most of the black letters are there.

Exactly! With all his talk of "energy" and "correlation", Joofa is making a huge meal out of stating the bleedin' obvious:

• An image being an array of positive numbers, it has a non-zero average, which appears as a high value at (0, 0). Shifting it to zero mean allows you to see the frequency distribution more clearly.
• The majority of people aren't photographing grids of regular lines or huge checkerboards - a typical feature we identify visually in a photograph has a wide frequency spectrum.

Sorry, you have little idea what you are talking about. I didn't even mention a single location of (0,0). And, if you just restrict to that location, you don't even have a natural image as that would just mean a uniform gray patch.

Did you even bother to have a look at the image I posted? Was the central white blob isolated to just (0,0)?????

unbelievable stuff you guys come up with. And then project on to me .

Their point is that the value of the Fourier transform of something at the origin (0,0) is equal to the sum/integral of the signal. If the signal has nonzero mean, the value at the origin is inflated relative to the rest of the spectrum.

Yes, I know. But, that hasn't much to do with the issue at hand. The white blob is not a single point.

-- hide signature --
Complain
Post ()
Keyboard shortcuts: