Bangla Handwritten Numeral Character Recognition Using Directional Pattern

Talha Ibn Aziz, Al Shahriar Rubel, Md Sirajus Salekin, Rafsanjany Kushol

Abstract: Handwritten character recognition has become a challenging and interesting field in the recent days due to its complex character shapes and huge pragmatic applications. A lot of research is already done and underway on English alphabets and numerals recognition. But in case of Bangla, even being the fifth largest spoken language in the world, has not undergone that much research. Besides, different complex shape of Bangla character makes the recognition more challenging. In this paper, we propose a directional pattern approach for feature extraction of Bangla numeric characters which attains a high accuracy of recognition. We use Local Directional Pattern (LDP) and Gradient Directional Pattern (GDP) for feature extraction and then two well-known machine learning algorithms, K-Nearest Neighbour (KNN) and Support Vector Machine (SVM), to classify the numeric character. We also ensemble the pattern oriented results to enhance the accuracy. Experimental results on the benchmark dataset CMATERdb 3.1.1 demonstrates an astounding recognition rate of accuracy 95.62% without preprocessing the data.