Handwritten Digit Recognition using ML

DOI

No DOI

Published In

Volume 3, Issue 3 (2022)

Published Online

May-June 2022

Pages

78-82

Article Type

Research Article

Check for updates

No DOI available for cross-reference

Abstract: Handwritten digit recognition (HDR) is the detection of digit from images, documents, car number plates and other sources and changes them in machine-readable shape for further processing. The accurate recognition of intricate-shaped compound handwritten digit is still a great challenge. Recent advances in convolution neural network (CNN) have made great progress in HDR (Handwritten Digit Recognition) by learning discriminatory characteristics from large amounts of raw data. In this paper, CNN is implemented to recognize the digits from a tested dataset. The main focus of this work is to investigate CNN capability to recognize the digit from the image, documented dataset and the accuracy of recognition with training and testing. CNN recognizes the digits by considering the forms and contrasting the features that differentiate among digits. Our CNN implementation is experimented with the dataset MNIST to obtain the accuracy of handwritten digits. Test result provides that an accuracy of 93.90% accuracy is obtained on 250 images with a training set of 1000 images from MNIST. The aim of this work is to review existing methods for the handwritten digit recognition problem using machine learning algorithms and implement one of them for a user-friendly web application. The main tasks the application provides a solution for are handwriting recognition based on touch input, handwriting recognition from live camera frames or a picture file, learning new characters, and learning interactively based on user's feedback on written format. The recognition model we have chosen is a multilayer perceptron’s, a feed forward artificial neural network (ANN), especially because of its high performance on nonlinearly separable problems. It has also proved powerful in OCR and ICR systems that could be seen as a further extension of this work. We had evaluated the perceptron's performance and configured its parameters in the Python programming language, after which we implemented the Web application using the same perception architecture, learning parameters and optimization algorithms. The application was then tested on a training set consisting of digits.