Forums

Newbie question

Started by Skice November 26, 2005
I have a 2d signal (an image, actually) of which I want to calculate
ACF.
I tried to use the IFFT/FFT approach, but when I rotate the image I
notice some strange effects (something aliasing-like). So I tried to
pad my image both vertically and horizontally with zeros, doubling its
dimension before calculating the FFT. Am I doing right? Someone can
help me? Thank you

Lorenzo

"Skice" <skice83@gmail.com> writes:

> I have a 2d signal (an image, actually) of which I want to calculate > ACF. > I tried to use the IFFT/FFT approach, but when I rotate the image I > notice some strange effects (something aliasing-like). So I tried to > pad my image both vertically and horizontally with zeros, doubling its > dimension before calculating the FFT. Am I doing right? Someone can > help me? Thank you
Yes, you're doing the right thing. The issue is that when you convolve any signal of length N with itself (time-reversed), which is what happens when you calculate the autocorrelation function (ACF) is that the result is of length 2N-1. This happens regardless whether the signal is 1D (e.g. audio), 2D (e.g. images), 3D (e.g. video) or higher dimensional. Ciao, Peter K.