An ANN seems overkill to me.

In the images, you have at least two different fonts (for those who cannot read docx files, rename it to zip, and you will find ten jpg and one png image).

Are the images calibrated in the sense that the height of a digit is always the same number of pixels? If yes, you are lucky, if not, you probably need to calibrate them or include magnification as an extra variable to optimize.

I would create a set of ten images for each font with uniform background, calculate an Euclidean distance-transform to the edges, and use these to do chamfer matching.

That method is fast and pretty tolerant to variably backgrounds.

Nice problem and good luck!

@Lambert Zijp, nice tip on the rename to zip :)

@Paul Strugess, can you give me any suggestions pertaining to my project?

We solved similar problem. Instead of credit card we had metal ingot - as same as in your case we had background with same color as the letters. Its similar to CAPTCHA problem. The paper is available at http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6608491&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D6608491

If you can, we can colaborate on the problem, we can test your input with our sofware etc.

Hello Kumar, Please g through these points ok,

A device for interacting with rendered documents includes a scanner and memory. The device can include a small, stand-alone scanning device, or it may include at least part of another device, such as at least part of a pen-style device, at least part of a mouse, at least part of a cell phone, at least part of a remote control, or at least part of a personal digital assistant, among numerous possibilities. The device can include one or more means to establish communication with at least one of a separate computing device or a network, and can be used both when such communication is currently established (“on-line”) and also in the absence of any such established communication (“off-line”).

Information from rendered documents is scanned to the memory, and a user of the device is signaled when sufficient information has been obtained to determine at least one action associated with the scanned information. For example, the user may be signaled when sufficient information has been obtained to determine at least one document comprising the scanned information, or the user may be signaled when enough information has been has been obtained to determine a commerce operation, an annotation, a format operation, or identifying an advertisement.

Determining when sufficient information has been obtained may involve ascertaining a type or types of the scanned symbols, counts or relative positions of the scanned symbols of various types, an overall number of scanned symbols, a number of scanned words, a count of stop and non-stop words or symbols, or the frequency of occurrence of words or symbols of various types, among other things. The methods used for making such a determination may differ depending on whether the device is on-line or off-line.

Scan context information may be stored in association with the scanned information in the memory. The scan context information may facilitate determination of an action or actions using or otherwise involving the scanned information.

Subsequent to the scanning, upon establishing communication between the device and at least one of a separate computing device or a network, the scanned information may be transferred to at least one of the separate computing device or the network, for further processing and to carry out actions using or involving the scanned information. Instances of scanned information may be accumulated in a repository for the user, and may be distinguished according to numerous criteria, including, for example, whether the instance was collected on-line or off-line, whether the system has presented the user with one or more possible actions that may be associated with the instance, and whether the user has indicated one or more actions to be performed with respect to the instance.

One of my undergraduate students group did similar work around 9 years back: optical character recognition to recognize car chassis number (which is embossed on metal and hence font is same colour as the background) using ANN. We published a paper on that work a little later. Go through it and share your queries.

Parul Shah, Sunil K., Taskeen N., G. Nikita, K. Kaushik, L. Ketan, “OCR-based Chassis-

Number Recognition using Artificial Neural Networks”, Proc. of IEEE International

Conference on Vehicular Electronics and Safety (ICVES), Pune, India, pp. 31 – 34, Nov.

2009.

https://www.researchgate.net/publication/224109006_OCR-based_chassis-number_recognition_using_artificial_neural_networks?ev=prf_pub

Conference Paper OCR-based chassis-number recognition using artificial neural networks

Parul Nilesh Shah : Thanks for your response. Please check your inbox.

@Kumar Abhishek " can you give me any suggestions pertaining to my project?"

Sorry, i don't have any experience with credit card scanning (i hope it is for a legitimate application btw), but if I had to do it, I would probably try to adapt a number plate recognition system, as it seems to be kind of similar to me at first glance:

There is an outline of such a system in:

Mastering OpenCV with Practical Computer Vision Projects

http://www.amazon.co.uk/Mastering-OpenCV-Practical-Computer-Projects/dp/1849517827

With code here:

https://github.com/MasteringOpenCV/code/tree/master/Chapter5_NumberPlateRecognition

Cheers

We made digits recognition for orientation detection of small documents including credit cards. You can find possible way for solving your problems in the paper "Intellectual Two-sided card copy" http://www.gc2011.graphicon.ru/files/gc2011/proceedings/conference/gc2011Safonov.pdf