Prev: Mrdivide problem in matlab
Next: C.Net Assembly interface and Methods
From: sudipta pramanik on 12 Aug 2010 09:06
I have two fourier spectrum image generated by fft transform.These two images are certainly different.Can any body how to differentiate the two images using some parameter?
From: Arthur Thompson on 13 Aug 2010 17:18
"sudipta pramanik" <pramanik_03(a)yahoo.co.in> wrote in message <i40rjt$c9d$1(a)fred.mathworks.com>...
> I have two fourier spectrum image generated by fft transform.These two images are certainly different.Can any body how to differentiate the two images using some parameter?

If I may re-state your question: Given two different images, and two different fourier transforms of images, how can you intuitively match each image to its fourier transform? If that is your question, I recommend looking at some example transform pairs at, for example, this site:

http://www.cs.unm.edu/~brayer/vision/fourier.html

If not, can you re-state your question?
From: Walter Roberson on 13 Aug 2010 17:47
Arthur Thompson wrote:
> "sudipta pramanik" <pramanik_03(a)yahoo.co.in> wrote in message
> <i40rjt$c9d$1(a)fred.mathworks.com>...
>> I have two fourier spectrum image generated by fft transform.These two
>> images are certainly different.Can any body how to differentiate the
>> two images using some parameter?
>
> If I may re-state your question: Given two different images, and two
> different fourier transforms of images, how can you intuitively match
> each image to its fourier transform? If that is your question, I
> recommend looking at some example transform pairs at, for example, this
> site:
>
> http://www.cs.unm.edu/~brayer/vision/fourier.html
>
> If not, can you re-state your question?

That wasn't my interpretation at all. My interpretation was:


Given two fourier spectra vectors, compute a _single_ measurement
("parameter", singular) that can be used to compare the "similarity" of the
spectra, with one value of the measurement reflecting identical spectra and
measurements that come out further from that key value reflecting increasing
dissimilarity. This measurement must measure "similarity" even if the two
spectra have different number of points {*}"

{*} This requirement that the measure ("parameter") must work for spectra with
different number of points can be deduced from the fact that the original
poster did not specify that the two spectra were the same size.


My alternative parsing is that we are perhaps not given the fourier spectra
vectors themselves, but only _plots_ of the vectors -- possibly only the real
or imag or abs parts at that; these plots would not be required to be to the
same scale, and it is not clear whether we would have access to the handle
graphics representation of the plots or only to screen-captures of the plots.



I did not answer the original poster because I am at a loss to figure out how
a _single_ scalar measurement ("parameter") could be used to decide the
similarity of two different two-dimension plots without there being a serious
risk of false positives or false negatives.
From: Greg Heath on 13 Aug 2010 19:51
On Aug 12, 9:06 am, "sudipta pramanik" <pramanik...(a)yahoo.co.in>
wrote:
> I have two fourier spectrum image generated by fft transform.These two images are certainly different.Can any body how to differentiate the two images using some
> parameter?

I would chose MSE as the most natural dissimilarity measure.

MSE21 = mean(abs(X2(:)-X1(:)).^2)

If neither X1 or X2 can be considered a preferred reference,

NMSE21 = MSE21/sqrt(MSE10*MSE20)

is a reasonable scale invariant normalization where

MSEi0 = mean(abs(Xi(:)-mean(Xi(:))).^2)


Hope this helps.

Greg
 | 
Pages: 1
Prev: Mrdivide problem in matlab
Next: C.Net Assembly interface and Methods