1

I was looking at all the approaches used to detect handwriting/Signature forgery. With the automatic feature matching algorithms like SIFT,SURF, is it possible to use these to detect handwriting forgery?

Also what are the ways to extract features from scanned images?

user987339
  • 10,519
  • 8
  • 40
  • 45
FelixAuror
  • 19
  • 3
  • Possible duplicate of [Is there any code or algorithm for signature recognition?](http://stackoverflow.com/questions/8907267/is-there-any-code-or-algorithm-for-signature-recognition) – Roger Rowland Feb 19 '14 at 12:44
  • I don't think that your decision to use SURF or SIFT is a good one. – Y.AL Feb 19 '14 at 15:14
  • yeah, SURF or SIFT maybe bad decisions since what's important you need to analyze the strokes. – azer89 Feb 20 '14 at 00:02

2 Answers2

1

Actually you have two objectives: - Handwriting / you need to use some machine learning (SVM or neural networks). - Signature / you may also apply the previous propositions, or some shape matching techniques.

To be able help you, please ascertain if you want to recognize or just detect?

Siddharth Kamaria
  • 2,448
  • 2
  • 17
  • 37
Y.AL
  • 1,808
  • 13
  • 27
  • I only want to detect whether it is a forgery or not, hence thought of using SIFT, but i'm not sure if that is algorithm that I need to be using. – FelixAuror Feb 20 '14 at 16:56
0

From a quick experiment SURF isn't that great for handwriting recognition. Comparing the feature vectors computed by SURF for these two images from the IAM Handwriting Database, you can see that only 1 keypoint is matched (shown by the line between the two images). A number to strive for would probably be 50 matched keypoints.

SURF on two images of the same word

drhoffma
  • 1
  • 2