Online Handwritten Kannada Character Recognition and Translation to English

Abstract

Abstract—The objective of the paper is to develop a technique that can efficiently recognize hand-written characters of Kannada language and then translate the same to English language. Online recognition refers to the processing of the characters as and when it is written on digitizer and recognizes the class which it belongs to. The iball 5540U Pen Tablet is used to collect the handwritten character samples and to build the database. The sensor picks up the pen- tip movements and also the pen-up/down switching. Kannada script consists of 6,23,893 characters. The complexity is reduced from 6,23,893 to 167 by segmentation approach. 50 samples of each character totaling to 8350 samples are used to build the database. Pre-processing technique includes smoothing, re-sampling and normalization. Spatial domain features such as normalized coordinates, normalized trajectory, normalized deviation are extracted from character samples. The efficiency of character recognition is improved by using various machine learning algorithms.  K-NN with K=1 and 3, SVM, Naïve Bayes Classifier are used to recognize the characters. Simulation result reveal a recognition rate of 78.41% using Normalised Coordinate features and SVM. Word translation rate is 92% for commonly used 50 proper nouns with different number of strokes.

Key Words: Strokes, Normalized Coordinates, K-NN, SVM, Bayes Classifier.

Cite this Article

Indira K, Geetha R. Online Handwritten Kannada Character Recognition and Translation to English. Journal of Image Processing & Pattern Recognition Progress. 2019; 6(2): 37–53p.