I have the problem that I need to recover from a photo an object.
I have for different objects all in all 5 categories and I need to check: In which category belongs the object of a new photo?
If I shoot now a photo, the program compare it to other images and then say:
"Bingo, this is the same object like in the database.
The database say for the old photo category 3, so it's is category 3."
The problem is now, how can I compare images if the objects are not exactly the same (pixel-for-pixel)?
But I have also some positive news:
The object will be in a limited area(30x30cm), with a good lightning (So if you have a red apple, next time the will have the same red) and the camera is always on the same position.
The only problem is, that the object could rotate. The photos will be made from above (bird's eye view).
I have some example images (I know that at the moment the lightning isn't the best(lot's of shadows) but that will be improved):
I tried imagemagick as that I made all images Black/White (I could crop out the background with help of the difference method and an image only from the background) and append them together in a big photo. I made another photo from one of the objects, made it Black/White and compare it but the process need a lot of time (like it's said in the tutorials section) and was very sensible (even with a -fuzz value) with different positions of the object... So it wasn't the right thing.
I've googled a lot and found something like co-occurrence matrices to calculate the pixels from a given object and the distances between them. In a paper from some students it looked very good, they could even recognize a rice package which was rotated in a shelf.
Do you have any suggestion what I can do to discover the object in a photo?
Regards and thanks,
TheGecko
