Interpretation of Handwritten Documents Using ML Algorithms

Abstract

Handwritten character recognition is a continuing field of research which covers artificial intelligence, computer vision, neural networks, and pattern recognition. An algorithm that executes handwriting recognition can acquire and detect characteristics from given handwritten document/image as input and convert them to a machine-readable form. It is also known as the task to convert the input text to extracting the features with the help of symbol representation and icons of each letter. The main goal of this handwritten text recognition is to identify an input character and text on a scanned image which are written in cursive writing, and these features are extracted with each input character pixels. Each character dataset contains 26 alphabets. IAM datasets are used for training the characters and for classification and recognition. The output is generated in the form of human-readable text. In this paper, the handwritten documents has been interpreted using machine learning algorithms such as connectionist temporal classification (CTC), long short-term memory networks (LSTMs), and generative adversarial networks (GANs). The results show the comparison of feature extraction based on proposed approach with convolutional neural networks (CNN) and recurrent neural network (RNN) with real-time datasets gain in better performance when using these algorithms.