Features to be considered in classifying currency images?

Presently, I am considering color and dimension of currency images as important features in classifying given huge set of currency images.

Is it necessary to consider few more features in order to increase the efficiency of classification?

What can be the other features?

Answer 1

[Anand]

use google: http://www.ijcsit.com/docs/Volume%204/Vol4Issue1/ijcsit2013040129.pdf

[Sushma]

Hello..Here they have considered ROI but i am applying feature extraction

Answer 2

[Image Analyst]

I think color and size should be enough. If the notes are very similar to each other, like US notes, then you can look at the colors in smaller areas, like you divide the note up into quadrants or into 10 by 10 tiles.

[Sushma]

I will try doing that

thanks

[Sushma]

Btw my image set consists of old and soiled currency notes as well..which may cause some color variation for a smaller area

[Image Analyst]

Yes. It's going to have to be robust enough to handle that. If you have a 10 by 10 grid over the note then you could check each one and each tile "votes" as to what kind of note it thinks it is. Then you call the whole note whichever type of note has the most votes. For example, you have 10x10 = 100 regions. 90 think it's a $10 note, 5 think it's a $5 note, and 5 regions think it's a $20 notes. So call it a $10 note because that's what had the most votes.