Enhancing Kannada Handwritten Text Processing: A Deep Learning Approach to Optimized Recognition and Segmentation

Digitalization ensures that information is available in diverse regional languages, empowering more cultures and perspectives to be heard and understood. One of the regional languages considered for empowering information access is the Handwritten Kannada document. Extracting text from these documents requires overcoming several obstacles, such as deciphering diverse handwriting styles, accommodating inconsistencies in character size, and the presence of multiple touches between characters. The present paper explored recognizing and segmenting Kannada handwritten characters using a deep learning model, specifically YOLOv8. While YOLOv8 is primarily known for real-time object detection, the paper suggests its potential for character detection tasks. The model achieved a promising mean Average Precision (mAP) of 96.8% at a threshold of 0.5 on a hybrid dataset consisting of 2476 images and 95.0% on character segmentation. This experiment adds to the growing body of research exploring YOLOv8’s capabilities beyond traditional real-time object detection and instance segmentation.