Volume 17, No. 4, 2020
Post-Processing Techniques To Enhance The Performance Of Gurmukhi Text Recognition System
Dr. Sukhdev Singh
Abstract
The development of the Gurmukhi Text recognition system is a challenging task as Gurmukhi text characters tend to intermix with other characters because of inherited structure. Particularly when the machine is trained to extract and recognize Gurmukhi text from natural science, the problem turns into many folds. The natural scene images carry complex backgrounds which intermixed with text regions and raised various issues in recognition of text. The present piece of work discusses experiments carried out to improve the performance of the system developed for the extraction of Gurmukhi text from natural scene images. The study revealed that dictionary-based error detection and correction is the best method to improve the results of the system. Using post-processing, the present system’s performance level raised from 92.0% to 94.5% for Gurmukhi alphabets and 95.5 to 98.5% for digits.
Pages: 219-223
Keywords: Gurmukhi text recognition, Post-processing, Error Detection, Dictionary Lookup table search.