I'm making a simple OCR detector in MATLAB. For now, only for uppercase letters (A-Z).
I have got the algorithms done to segment lines and individual characters. Now, I need to compare each extracted character against a database of all the alphabet letters.
For that, I have used the projections of the binarized (black-and-white) character on the x-axis and y-axis.
To compare the two projections with the projections in the database I have used the cross-correlation (with the mean substracted, to remove the DC).
This works perfectly when the characters are exactly the same as those in the database, but when I extract the characters from another image, the characters will be rescaled, compressed... but the patterns remain, my algorithm is not useful, and worked only for 40-50% of the characters of the image.
I put an example:
This are the projections of the character C of the database (on the top) and of the character D of the image (on the bottom).
And this are the projections of the character D of the database (on the top) and of the character D of the image (on the bottom).
Ok, then, my code says that the projections of C are more resamblance than those of D. With numbers, the correlation of C is 0.9 for horizontal and 0.7 for vertical. And the correlation of D is 0.78 for horizontal and 0.67 for vertical.
I am going to attach my code, the final question is: What is the best algorithm (not too advanced) that could work well?
My code:
proj_horizontal_database = caracter_alfabeto(2, :) - mean(caracter_alfabeto(2, :));
proj_vertical_database = caracter_alfabeto(1, :) - mean(caracter_alfabeto(1, :));
proj_horizontal_letter = sum(caracteresLeidos{i}{j}, 1) - mean(sum(caracteresLeidos{i}{j}, 1));
proj_vertical_letter = sum(caracteresLeidos{i}{j}, 2)' - mean(sum(caracteresLeidos{i}{j}, 2)');
correlation = xcorr(proj_horizontal_database, proj_horizontal_letter, 'coeff');
correlation_horizontal = max(correlacion);
correlation = xcorr(proj_vertical_database, proj_vertical_letter, 'coeff');
correlation_vertical = max(correlacion);
