Offline Handwritten Amharic Character Recognition

Offline handwritten Amharic character recognition aims to automatically translate handwritten Amharic text into digital form, addressing the scarcity of digital resources for this low-resource language. Current research focuses on adapting deep learning models, such as convolutional and recurrent neural networks with connectionist temporal classification (CTC) loss, often incorporating techniques like few-shot learning to overcome data limitations. Exploiting inherent structural similarities within the Amharic alphabet, through auxiliary tasks or data augmentation, is a key strategy to improve recognition accuracy. This work is significant for expanding OCR capabilities to under-resourced languages and facilitating broader access to Amharic literature and information.