Devanagari Handwritten Compound Character Recognition Using Various Machine Learning Algorithms

Handwriting is unique for each person. The formation of letters, compound characters, spaces and numbers by each person is so unique that handwriting is considered as legal evidence in forgery cases. Hence accurate recognition of handwritten characters holds huge significance. Especially compound characters are differing in writing ways and there are some characters which can be written in many ways so it is complex to recognize. Segmentation is another issue occurs due to structure of the character is very complex. This paper proposes a method to recognize compound handwritten Devanagari characters using several machine learning algorithms like K-Nearest Neighbours (K-NN) and Support Vector Machine (SVM). Techniques used for extracting features from the images are playing a significant role in the identification of compound characters. In this paper, the Zoning feature extraction technique is used combined with other features like eccentricity, equivalent diameter and angle. We created a new dataset of 45 compound characters. Using 10 cross-validation we achieved the highest 96.62% accuracy by SVM classifier with training time 1.56 seconds.