Optical Character Recognition using Convolutional Neural Network

Abstract

The Optical Character Recognition is used to detect the information from natural images and convert them into words, symbols and numbers. Known as computer vision, this is a sub-field of image processing that deals with the “sight and recognition” by a machine. The method implemented in this study is done using a concept known as Convolutional Neural Networks (CNN). This is a Deep Learning algorithm that can take in an input image, assign importance (learnable weights and biases) to various aspects/objects in the image and be able to differentiate one from the other. CNN used for recognition takes an input image, pre-processes it and recognises it. The CNN algorithm, built from a limited dataset, can produce results suitable for small scale implementations with limited storage space. The chapter begins with creating a dataset consisting of images for text and symbols. The dataset is divided into training, testing and validation data. CNN is the main algorithm used from a training stand point and can provide accurate results. An input image is loaded as a file or using a webcam and fed into the developed OCR system. It will then undergo pre-processing (binarisation, equalisation, reshaping, resizing), one hot coding, creation of neural layers and finally, the output image can be recognised.