picture of picture detection

[Bharath K]

I am working on detecting whether picture is original picture or picture of picture. But my results are not up to the mark. I am brand new to image processing and I would like some suggestions to move forward.

I have tried following things:

1) Extracted features using mahotas library(i.e texture of the image)

2) Built SVM classifier and logistic regression.

But I dont think all features of mahotas could be useful to detect picture of picture.

Can anyone tell me what would be major difference between original picture and picture of picture?

Any suggestion would be great help for me..

Thanks

Bharath

[Luis Pedro Coelho]

I guess the first questions are whether you have training data and
whether it is easy for humans to distinguish the two categories.

If the answer is yes, then the problem should be doable
computationally...

Cheers,
Luis

Luis Pedro Coelho | EMBL | http://luispedro.org
My blog: http://metarabbit.wordpress.com

> --
> You received this message because you are subscribed to the Google Groups
> "pythonvision" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to pythonvision...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Bharath K]

Thanks Luis for the quick response.

Yeah, I do have training data. Not many but to start with I have 50 Normal and 50 Picture of picture images.

I have gone through http://haralick.org/journals/TexturalFeatures.pdf and used mahotas to get features. But for this problem I dont think I need to use all the features giving my mahotas.

With 10/10 Data, it gave me 80% accuracy but I feel I am missing something.

Could you direct in right direction?

Thanks

Bharath