you have to look at some very good Image processing algorithms and do the preprocessing and use some pattern recognition principles, then may be if needed you can use neural networks. neural networks is always not the ultimate right choice.
i want to apply it into a pocket product(scanning pen). so hoping this algorithm is simple and can be executed by MCU. whether i need think over the speed of sanning? because everyone will scan book at defferent speed.
Thre is a product that has a pen shap that scan text.
The main functionality is to scan and translate text. the product name is Quicktionary - the company is located in Israel.