A Machine Learning Approach for Bengali Handwritten Vowel Character Recognition

Abstract

Handwritten character recognition has numerous complexities because of the various shapes and numbers of characters. There are many manually written character recognition strategies proposed for the English language as well as for other significant dialects. Bengali is generally considered the fifth most spoken local language on the planet. It is the official and most broadly spoken language of Bangladesh and the second most generally talked amongst the 22 booked dialects of India. To improve Bengali Handwritten Character Recognition, we developed a different approach using face mapping in this study. It is quite effective in distinguishing different characters. The actual highlight is getting more efficient recognition results than expected using a simple machine learning technique. The proposed methodology utilizes the python library Scikit-Learn comprising NumPy, Pandas, and Matplotlib alongside Support Vector Machine (SVM). This proposed model uses a dataset derived from the BanglaLekha-Isolated dataset for the training and testing parts. The new approach presents positive results and looks promising. It showed accuracy levels up to 94% for a specific character and 91% on an average for all the characters.