Optimized Handwritten Digit Recognition: A Convolutional Neural Network Approach
Ankur Rai, Siddharth Singh, Sumit Kushwaha · 2024
This Regardless of the fact that image analysis has always been a topic of research, it continues to capture the interest of many researchers. Recognizing the hand-written character is among the most practical difficulties in pattern recognition applications because there are so many different writing patterns, hence automatic recognition and classification is very challenging. Recognition of digit is being used in a several applications such as automatic scanning of bank cheque, postal address, postal mail sorting, data entry forms, and a lot more. The core problem of digit recognition lies within capacity to design an optimized solution that can recognize handwritten numerals which is provided by users via a tablets, scanner, and other digital devices. However, because of time constraints or because the model is designed to a certain purpose, several articles opt to compare models only on one or two datasets. For handwritten digit recognition, previous approaches missed high accuracy and processing speed. The moto of this research is just to implement a model of Convolutional Neural Network (CNN) that is both simple and correct in the direction of classifying or grouping digits written by hand for number of data sets. With two datasets, the researcher presented five unique CNN architectures for training and testing. Dataset-1st has 12000 data of MNIST, while dataset-2nd contains 29400 data of Kaggle. The suggested model of CNN first took out features before performing classification. The performance of these CNN models was enhanced using the Stochastic Gradient Descent with the Momentum (SGDM) optimizer.