A Deep Learning Approach for Recognizing Bengali Characters and Numbers

Abstract

Bengali is one of the world's most widely spoken languages. But it is a challenging issue to recognize Bengali Alphabet and Numbers from handwritten documents. It has 50 fundamental alphabets and 10 Fundamental Numbers. Because of their diverse sizes, similarities between alphabets, and distinct writing methods, recognizing the Bengali handwritten Alphabet and Numbers is a difficult task. In this research, a deep learning-based approach is proposed combining CNN-BiLSTM for recognizing Bengali Alphabets as well as Numbers and the classification is followed by a multi-support vector machine (M-SVM). We introduced a dataset named Bengali Handwritten Alphabet and Numbers (BHAN-2022) dataset which was applied to evaluate the perfection of the proposed system. A Morphological Operation is also used before testing the input image which assists the model to remove the unnecessary part from the input image. The experimental findings show that the suggested model recognizes Bengali handwritten Alphabets as well as Numbers with an average accuracy of 97.08%.