Abstract
Character recognition is the process that allows the automatic identification of character images, which is generally referred as Optical Character Recognition (OCR). The characters are either handwritten or typed. This study proposed a novel OCR approach based on the likelihood functions of pixels, which were obtained by averaging a trained set of character images. A Bayesian fusion process for all pixel probabilities decides the recognition of characters. Further tests using Support Vector Machine (SVM) classifier were carried out on characters with the same shape. This method was used to test noisy images and achieved an accuracy of 97.95%, thus, outperforming other OCR methods.
Original language | English |
---|---|
Pages (from-to) | 2319-2325 |
Number of pages | 7 |
Journal | Journal of Applied Sciences |
Volume | 12 |
Issue number | 22 |
DOIs | |
Publication status | Published - 2012 |
Keywords
- Bayes theorem
- Fusion
- Optical character recognition
- Pixel likelihood function
ASJC Scopus subject areas
- General