About the localization and feature extraction in bank notes recognition

I am doing a project about the recognition of the serial numbers in bank notes. After I finished the gray scale normalization of the image, I don't know how to extract the serial numbers from the image. Also, I need to extract the features of each character for the comparing with the serial numbers. However, may I know if there any methods to do this extraction. Thank you very much and Happy new year.

Answer 1

[Image Analyst]

Use the Computer Vision System Toolbox to do OCR: http://www.mathworks.com/help/vision/ref/ocr.html#bt548t1-2_1

[YIP]

May I know if there any method other than using the computer vision system toolbox?It is because I need to do the project in R2012a version. So may I know if there any code for OCR?

[Image Analyst]

No, the only "built-in" OCR is in the Computer Vision System Toolbox. If you don't want to use that then you'll need to find an alternate, like in the File Exchange, or write your own.

[Image Analyst]

What is the question?

[YIP]

May I know if the file"new template", which is attached, is made by the COMPUTER VISION SYSTEM TOOLBOX? May I have any method to read the code in it?

[Image Analyst]

Sorry - I don't have that toolbox right now so I can't test your file. The mat file itself was probably made with the save() function but the variables in it may have come from a toolbox function - I don't know.

[YIP]

I can load the file, but I don't know how it is generated. May I konw how it is generated? Is it related to create_template.m??

[Image Analyst]

.mat files are created by calling the save() function.

[YIP]

Thank you. Now I have another question. May I know how to make the bitmap figure in the tempalte( characters file)? I try to make the new temaplates by copying the bitmap figure after the program cut the [42 24] character out, but I cannot only cut out the character out without the blackground. May I know if I am right to do these procedures to extract the template figure??

[Image Analyst]

An image must remain rectangular, so you'll need to have both white foreground pixels (representing the M) and the black background pixels. If you want, you can use find() to get a list of the rows and columns where the white pixels reside, but that's just a list of coordinates, not an image.