Comparative Analysis of CNN and RNN Models for Bangla Hand Written Character Recognition

Bangla, being the fifth most spoken language globally, presents unique challenges in character recognition due to its complex script structure and modifiers. This paper explores the comparative efficacy of Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) in Bangla character recognition [1], utilizing a comprehensive dataset of 15000 handwritten Bangla characters that includes consonants, vowels and modifiers. Our work uniquely addresses the challenges of recognizing complex Bangla script structures such as overlapping modifiers and compound characters, which have not been thoroughly explored in prior studies Our analysis highlights CNN's spatial feature extraction capabilities, achieving 96.46% accuracy, and the RNN's sequential dependency modeling, which yields 82% accuracy. We evaluated both models on metrics such as accuracy, execution time, and computational cost, inspired by advances in hybrid models and horizontal zoning techniques. This work contributes to the growing field of optical character recognition (OCR) for underrepresented languages, demonstrating key strengths and limitations of CNN and RNN models for Bangla scripts