Optical Character Recognition system with Projection Profile based segmentation and Deep Learning Techniques
Arockia Rexi F, Lija Jacob · 2022
Optical character recognition is the solution to convert text from printed or scanned documents into editable data. This project is aimed at building a Optical character recognition system that recognizes digital text. A document is first detected using contour-based detection technique without altering the angle of the image and is segmented into lines, once the lines are segmented the words embedded in them are extracted. This segmentation is done using projection profiling method. Characters are then segmented words with vertical projection profiling from the extracted words. These characters are fed into an image recognition model for recognition. The recognition model is CNN based deep learning model. Modified VGG16 architecture is used here to extract maximum features from the images and then classify them. To train the model a dataset is created from a repository of digital character dataset. The dataset consists of images of 153 font variants.