Hi,
These are my two cents as a Computer Vision/Image processing(Biometrics :
in the past) researcher.
1. I do not have a list of softwares for the specific function and am not
sure if you can find one for "free" (free as in freedom or free beer).
2. If you want to built one, here are my suggestions. This is what I would
do. It might take sometime (say a week or so) to get it working depending
on your constraints (or accuracy) and your knowledge of the area.
a. First and foremost, you will have to create a bounding box of where
to find the signature.
b. Once that is done, the simplest method would be do a simple (sum of
differences), which is to compare both images locations and see if they
have same pixel intensities.
c. Now get the score for the whole image and anything above a threshold
would be the right signature and below would not be.
d. This will only work, if the signature was exactly the same or very
very similar and both have the same location inside the bounding box,
meaning if it is not translated (no lateral movement) and the inter
character difference is also similar. So the truth is that, it won't work
most of the time.
e. Now a good hack of that would be to do this. I haven't tried this,
but feel it might be a good experience. You could give this as an input to
OCR recognition software to detect this as a single character, the
character being the signature here. Many of the OCR softwares are trained
in such a way that, it does not care whether the input is translated or
not. One of the best results are using Yann Lecun's convolutional neural
networks : http://yann.lecun.com/exdb/lenet/index.html. But even simpler
ones based on SVM or even a Naive Bayes classifier should work.
f. So simply put, the idea is to train the software using sample of each
persons signature as the character to be recognized. Now during test if the
match is high, it is good match or else its not.
This is really a very coarse approach, based on 15 min thought.
Regards,
-Sumod
On Mon, Feb 6, 2012 at 6:28 AM, Nataraj S Narayan <[email protected]>wrote:
> Hi Srinivas
>
> What kind of algorithm does opencv use to compare images? Hand
> signatures by the same person might never look exactly similar, except
> that the pattern might be the same. Hope opencv does more of a
> pattern analysis.
>
> Where can I get some info regarding what goes behind handwriting or
> signature analysis? Modern banks like Icici might be using some kind
> of automation for this purpose instead of person checking out on the
> signature pattern manually?
>
> Warm regards
>
> Nataraj
>
> On Mon, Feb 6, 2012 at 12:46 PM, Shrinivasan T <[email protected]>
> wrote:
> > we can use opencv to compare images
> >
> > On Feb 6, 2012 12:35 PM, "Nataraj S Narayan" <[email protected]>
> wrote:
> >>
> >> Hi
> >>
> >> Anybody aware of any Linux based software that compares two
> >> signature images and
> >> checks if it is the signature of the same person?
> >>
> >> I mean any software that detects forged signatures?
> >>
> >> regards
> >>
> >> Nataraj
> >>
> >> _______________________________________________
> >> Indian Libre User Group Cochin Mailing List
> >> http://www.ilug-cochin.org/mailing-list/
> >>
> http://mail.ilug-cochin.org/mailman/listinfo/mailinglist_ilug-cochin.org
> >> #[email protected]
> >
> >
> > _______________________________________________
> > Indian Libre User Group Cochin Mailing List
> > http://www.ilug-cochin.org/mailing-list/
> > http://mail.ilug-cochin.org/mailman/listinfo/mailinglist_ilug-cochin.org
> > #[email protected]
>
> _______________________________________________
> Indian Libre User Group Cochin Mailing List
> http://www.ilug-cochin.org/mailing-list/
> http://mail.ilug-cochin.org/mailman/listinfo/mailinglist_ilug-cochin.org
> #[email protected]
>
_______________________________________________
Indian Libre User Group Cochin Mailing List
http://www.ilug-cochin.org/mailing-list/
http://mail.ilug-cochin.org/mailman/listinfo/mailinglist_ilug-cochin.org
#[email protected]
