2

We have a for fun project which require us to compare two black and white bitmaps of two signature and say whether they are the same persons signature. As this is just two loaded bitmaps rather than data captured from a tablet the approach is going to be a little different to normal signature recognition.

I am thinking it would require the following steps

  • Crop the bitmaps to just the signature
  • Try to work out some kind of rotation to align them
  • Resize to make the cropped / rotated bitmaps the same
  • Analyse the signature inside (maybe by breaking down into a grid)

Does anyone have any thoughts on this project? How to best do the rotation, comparison etc? Seen anything similar?

Matthew Hood
  • 958
  • 3
  • 13
  • 22
  • Will there be any fuzzy logic involved, or is it going to be either exactly the same (albeit rotated/scaled etc...) or completely different? – keyboardP Dec 01 '09 at 15:44
  • It will be fuzzy, its signature verification so will be "how similar are these two signatures" ie can we accept this as a signature done by the same person. – Matthew Hood Dec 01 '09 at 15:47
  • 2
    What research have you done so far? There are probably several armies of grad students working on this topic right now... – Jon Seigel Dec 01 '09 at 15:55
  • Off the top of my head, it's not too complicated. A simply neural network would handle this for you, whilst allowing fuzzy logic to take place. I'd go down the Self-Organizing Map route, and have the input vector include the signature's bitmap structure. Obtaining the correct input vector would be the trickiest part, so it really depends on what you have in mind/what you've already done. – keyboardP Dec 01 '09 at 16:22
  • Ive done a fair amount of looking around, most signature algorithms are based on a tablet so take into account position, time and pressure. This is more along the lines of image comparison so was looking for ideas on which approaches would more suit line image comparison. Been looking at things like HMM etc as I say it is just a for fun project. – Matthew Hood Dec 01 '09 at 16:51

1 Answers1

1

You may want to look at:SOMs for interesting pics (:D) as well as an example of how to compare image similarities.

There are two main types of Neural Networks - supervise and unsupervised. SOMs are unsupervised. Depending on your situation, you might want to take a look at supervised Neural Networks NNs are common, and quite straightforward to implement for the most part.

keyboardP
  • 68,824
  • 13
  • 156
  • 205