IMPROVING THE ACCURACY OF VIETNAMESE OPTICAL TOUCHING AND BREAKING CHARACTER RECOGNITION

Abstract

This paper propose an efficient method for improving the accuracy of Vietnamese optical touching and breaking character recognition. Basically, the propose method focus on three main step: i) improving the accuracy of character classification algorithm; ii) Determining the optimal set of the cut candidate; iii) Optimizing the searching the best result from cut candidate.  The performance of this method has been verified on three Vietnamese data sets, collected from reality with a total of 15270 lines of text, diverse in number, quality and font type. Experimental results show that this method has high accuracy and stability on experiment data sets and is fully capable of recognize poor quality input texts.