Transfer Learning Approach for Handwritten Character Recognition of Meitei Mayek Script Using Deep CNNs

Abstract

The digitization of low-resource scripts poses a significant challenge including structural uniqueness and limited computational support. This work aims to present an OCR System for the Meitei Mayek script using deep learning techniques, specifically transfer learning with the VGG16 Convolutional Neural Network. It uses a fine-tuned VGG-based convolutional architecture to perform the recognition of characters by modelling the spatial and geometric properties of handwritten symbols. The system uses a VGG-based convolutional architecture adapted for the Meitei Mayek characters, where convolution layers detect fundamental visual elements such as strokes, curves, intersections, and spatial orientations. Images are passed through successive layers in which low-level features are combined into abstract representations that describe the overall form of each character, while pooling operations reduce sensitivity to minor handwriting variations.