ABSTRACT
ABSTRACT: Optical character recognition (OCR) is the process of identifying characters in any form, with the help of a photo-electric device and computer software. Handwritten numeral recognition can be considered as a special area in OCR, where the characters are handwritten numerals. Unlike printed numeral recognition, handwritten numeral recognition is very difficult as different persons have different writing styles. Usually, Conventional OCR includes steps like pre-processing, segmentation, feature extraction and finally classification and recognition. Conventional OCR uses handcrafted features for feature extraction, whereas Convolutional Neural Networks (CNN) extracts features automatically. Hence by using CNN, we can eliminate the need for manual feature engineering. Recently many researchers have used deep learning architectures like CNN, Stacked or Deep Autoencoders and Deep Belief Networks for Handwritten Numeral Recognition and they have shown promising results compared to other methods. No one has yet implemented a system for recognizing Malayalam and Kannada handwritten numerals using CNN. So this is the first project to address Malayalam and Kannada handwritten numeral recognition using deep learning approach. Also, we compare the performance with another deep learning architecture, stacked autoencoder. The sole aim of this project is to automatically recognize and classify Malayalam and Kannada numerals using Convolutional Neural Network.
