Re: fft phase for object detection
- From: davem@xxxxxxxxx (Dave Martindale)
- Date: Fri, 20 Oct 2006 04:37:02 +0000 (UTC)
"Graham Jones" <x@xxx> writes:
Are you trying to say modulus and argument (or phase) when you say re and
im? Anyway that's the only way I can make sense of your question. I suggest
you try FFT-ing an image, dividing each value by its modulus (thus removing
the information contained in the modulus, leaving just phase information)
then inverse FTT-ing the result and looking at it. If the original image was
simple, it will probably still be recognisable.
Another interesting experiment: take two images the same size, call them
A and B. FFT each of them independently, and convert the output to polar
form (magnitude/phase angle). Now create a new complex array of the
same size as the FT of the first two, and fill each location with the
amplitude (modulus) from the corresponding location in A's FT and the
phase from the corresponding location in B's FT. (Essentially, you're
building a new composite FT whose amplitude information comes from one
source and whose phase comes from the other). Convert all the
coefficients from polar back to Cartesian form, and then do an IFFT.
The result, call it image C, will have features from both A and B mixed
together. But I'll bet it looks more like B than A. If that's true for
the set of images you try it with, you can say that the phase actually
carried *more* of the human-visible information in the image than the
amplitude did.
Dave
.
- References:
- fft phase for object detection
- From: richard
- Re: fft phase for object detection
- From: Martin Leese
- Re: fft phase for object detection
- From: richard
- Re: fft phase for object detection
- From: Graham Jones
- fft phase for object detection
- Prev by Date: Scale frame size..
- Next by Date: Re: Positive image interpolation
- Previous by thread: Re: fft phase for object detection
- Next by thread: Hough transform
- Index(es):
Relevant Pages
|