HANDWRITTEN DIGIT RECOGNITION USING MULTILAYER PERCEPTRONS AND BACKPROPAGATION LEARNING
Main Article Content
Abstract
Handwriting recognition has long posed a significant challenge in the realm of computer applications. However, the advent of neural networks has ushered in a new era, opening the door to a myriad of applications, including handwriting recognition, voice recognition, and complex decision-making through machine learning. In this project, we present a Java application that leverages neural networks to process image data, converting it into a 24 by 24 matrix pixel-wise. These data points fall within the range of 0 to 255 and are segmented digit by digit, subsequently organized into separate Excel sheets within a local storage-based Excel file.
To optimize performance, the dataset is further divided into two distinct subsets: a training dataset and a test dataset. The training dataset comprises 180 samples for each digit, totaling 1800 data rows for the training process. These pixel values are sequentially fed into a Back Propagate Neural Network implemented in Java 1.8, and training occurs using supervised learning principles. During training, we calculate errors for the expected outputs and make adjustments to the two weight matrices, sized at 784 by 200 and 200 by 10, to enhance the network's accuracy.