Abstract:
This research paper presents an in-depth
investigation into the application of Convolutional Neural
Networks (CNNs) for Tamil handwritten character recognition.
We explore existing research, methodologies, and cutting-edge
techniques, showcasing CNNs' effectiveness in achieving a
remarkable 95% accuracy. Our dataset comprises 247 Tamil
characters and 18 North Indian characters, accommodating
diverse writing styles. We tailor CNN architectures for Tamil
characters, implement advanced preprocessing, data
augmentation, and training methods to enhance model
performance. Our paper tackles challenges posed by accessible
datasets, offering remedies for data scarcity, class imbalance,
and writing style variations. Our distinct contribution lies in
achieving 95% accuracy across 247 Tamil characters and 18
North Indian characters, demonstrating CNNs' potential for
document processing, language preservation, and automation in
Tamil-speaking regions. This work advances the field by
introducing novel techniques, a comprehensive dataset, and
strategic insights, serving as a significant step forward in Tamil
character recognition.
