Character Recognition using Perpendicular Distance on Sweep Line and Chi-Square Statistic as classifier

Abstract

In ordered to identify an object in an image it is considered a single unit and this process is known as image processing. So, In this paper, we have proposed a novel feature extraction (FE) technique for character/digit recognition (CR) by applying perpendicular distance onto a sweep line from borders of the input character. Proposing a robust FE technique is crucial for any CR system for better performance. CR plays crucial role in many Image Processing (IP) applications. Before extracting the features of the image, process it by converting into grey scale and subsequently to binary image. A scan line is generated in the binary image and traversed perpendicularly from each point on the scan line to both directions to get the extreme end points which is taken as perpendicular distance. The extracted features are in a DB/text file for recognition of input characters. A data set containing 10, 000 images have been used for performance analysis and separated them into 2 different categories as training, testing sets and stored the extracted features in the DB/text file along with the label which it specifies while training and test the efficiency of the model. Chi-square statistic is used as classifier in this paper. We have achieved encouraging results using the proposed CR FE technique when compared with the other CR FE techniques from the literature.