## Projects

Research on Optical Handwritten Digit Recognition using Deep Learning Models on the MNIST Dataset

Hand written digit recognition is a fundamental problem, that still exists, in computer vision. The complexity arises due to variation in handwriting styles. This paper uses the MNIST dataset for handwritten digit recognition as it is a standardised benchmark. The MNIST dataset consists of preprocessed grayscale images that are normalised and centered based on pixel mass, reducing positional variance and background noise. Convolution neural networks such as VGG and ResNet are implemented and their experimental results are compared, demonstrating deep learning models achieve high accuracy and outperform conventional approaches. Performance is measured by accuracy and these metrics are tracked to show the strengths and weaknesses of the architectures. The insights these results provide are useful for helping us pick the right deep learning model for real world applications. This paper provides a comparative assessment of three deep learning architectures: VGG16, ResNet and Vision Transformer for optical handwritten digit recognition, using the MNIST dataset. The models' performance is evaluated using accuracy of classification, VGG16 provided an ccuracy of 98.53%, ResNet18 99.22% and Vision Transformer provided the highest accuracy at 99.33%. Research work published through IEEE, exploring applied computing and technology research.