Performance assessment of different image sizes for printed Gujarati and English digits using template matching

Abstract

This paper presents a system for separation and recognition of offline printed Gujarati and English digits using template matching. Sample images of different quality of papers were collected. They were scanned at 200 dpi. Various preprocessing operations were performed on the digitized images followed by segmentation. Segmented image of various sizes was normalized to get an image of uniform size. Then the pixel density was calculated as binary pattern and a feature vector was created. These features were used in template matching for the classification of digits. The recognition rate was tested on images of 3 different sizes viz. 24 × 24, 32 × 40, and 48 × 48 for offline printed Gujarati and English digits. We collected 200 image samples which include more than 4200 symbols of both Gujarati and English digits. The results were evaluated for different image sizes of 24 × 24, 32 × 40, and 48 × 48. The overall recognition rates were 97.43, 98.30, and 97.28 for Gujarati digits and 99.07, 98.88, and 99.34 for English digits respectively.