I think you may confuse something: OCR stands for "optical character recognition" - OCR algorithms are based on shape descriptors which assign shapes to letters by probability. Anything you do in order to assign text is some sort of an OCR algorithm. There are some basic shape features with which you coud try to recognize letters. The most crucial step is to segment single letters (handwriting is probably no option) - here some ideas to create trees that give you in the end probabilities for a binary object being any letter:
(1) symmetry in X or Y
"K" has symmetry in Y
"Y" has symmetry in X
"X" has symmetry in both
"P" is asymmetrical
(2) how many junctions (how many branches and loops) does the letter contain?
where do these junctions lay relative to each other
Straight lines vs. rounded lines
Attention: Serifs!
(3) how does the convex hull of a letter look like?
(4) there are tons of shape descriptors - some are easy to understand, some are less intuitive - e.g. hu- or zernike moments.
I suggest you read some basics literature about "OCR", "Shape Descriptors" and maybe "Decision Trees" and Classifiers, e.g. "Random Forest". A more sophisticated approach would be learned feature assignment based on a training set.
I read this following paper where they insist as "OCR gives good result for scanned document". And I need clarification about recognition process, by which technique text in images are printed.
Hello, it´s common to think that OCR is the only solution for characters identification but there are plenty algorithms that can achieve this task by example, haar, neural networks or so, character recognition is a matter of classification, so you may start from there.