Reading digits out of images possible?

In terms of the error level, in the early stages of programming this then you could just make it so that it aborts the attempt if it can’t figure out the numbers.

My bank has something with this facility: you can take a picture of a certain part of a bill and it will read in the digits. But if it can’t figure it out, then it will say so and you get the chance to enter them by hand.

@ignatz sorry buddy, but that was not me who made up that term, see for instance http://www.opticsinfobase.org/ol/abstract.cfm?id=12597 . Now i just put that term because:
1/ i want to show off :wink:
2/ if you google it you can find links to papers.
3/ they are really great for that kind of recognition: the set of resulting filters will really be the most efficient ones ever (best signal / to noise ratio). They are difficult to compute from scratch, but they probably have already been computed somewhere. If someone is interested i could compute them (at work) and post them here as an image. [edit] i must start from the picture of the numbers you want to recognize.

@Jmv38 - perhaps he should use Target-adaptive polarimetric synthetic aperture radar target discrimination using maximum average correlation height filters

(You see I can use long words too, the difference being I don’t understand them :-S )

@Ignatz, that gave me a good chuckle!

@Briarfox - to make it worse, that was a real paper I quoted there! I guess each time you improve the method, you get to add a couple of words on the end.

Mind Blow and here I thought you were being silly!

@Ignatz lol! :))
Target-adaptive polarimetric synthetic aperture radar target discrimination using maximum average correlation height filters will not be any help here:
1/ optical images do not have polarisation characteristics,
2/ radar images have a very specific noise structure (called ‘speckle noise’) that is not present in normal images, and so the optimal algorithm for radar are sub-optimal for normal images.

Curses, I thought I had beaten you there :((

@BriarFox - not sure if this thread is any use to you (http://twolivesleft.com/Codea/Talk/discussion/1350/simple-character-recognition-input-demo)

Also re - taking the original image and identifying the individual cells, have you thought about drawing a frame over the top of the camera input and then having the user move the iPad so that the meter digits line up to and fill up as much of the frame as possible?

That might make the process of finding the numbers quicker and easier.

That is a very smart solution @TechDojo to ease the computation load! Good point. =D>

Thanks @TechDojo that was my thought exactly. It would make it easier.