Training-Less Optical Recognition of Printed Characters Using Equation Fitting
|
 |
|
Post a Comment
|
|
|
|
|
CONTRIBUTORS:
|
|
|
JOURNAL:
|
|
|
YEAR:
|
2008
|
|
PUB TYPE:
|
Journal Article
|
|
SUBJECT(S):
|
optical character recognition; training-less; equation fitting
|
|
DISCIPLINE:
|
Engineering and Applied Sciences
|
|
HTTP:
|
http://www.ics.uplb.edu.ph/node/283
|
|
LANGUAGE:
|
English
|
|
PUB ID:
|
103-444-123
(Last edited on
2008/07/19 02:01:22 GMT-6)
|
|
SPONSOR(S):
|
|
|
ABSTRACT:
We present a novel computational method for optical recognition of printed characters based on our observation that a character image can be divided into several partitions, where each represents a simple mathematical equation. In this paper, we derived a set of mathematical equations to respectively fit parts of an alphanumeric character. Using these equations in tandem with digital image processing, we can identify characters without going through tedious supervised machine training sequences where large samples of learning patterns is required. We digitized ten samples each of printed alphanumeric characters, and store each image in a matrix M, where the (i,j)th matrix element mi,j = I+. To remove noise, we converted M into binary using a thresholding value h, setting all mi,j > h to 1, and to 0 otherwise. We “thinned” M by reducing to 0 some matrix elements in the Moore neighborhood of a chosen mi,j=1, following the Rutovitz and Hilditch thinning procedures, to make sure that a continuous thin line is visible in M. We divided the thinned M into six partitions: two columns and three rows. In each partition, we extracted the coordinates (i,j) of each mi,j=1 and subjected them to equation fitting. We determined which equation the image part fits best using a simple correlation analysis. We developed a table of patterns for each character, wherein a table entry that matches the 6-part pattern determines the alphanumeric character. We validated our method using 360 images where our method correctly classified 94% of Arial fonts. Of the 6% error, our method interchangeably identified C as G, 9 as S, and I as T. Based on the promising result from our validation method, we conclude that our method is a potential alternative optical recognition scheme, specifically on machine printed characters.
|
|
|
|
STATISTICS
|
|
Click on # to view
|
|
Citations
|
|
0
|
|
References
|
|
0
|
|
Comments
|
|
0
|
|
Quality
|
|
0/0.00
|
|
Interest
|
|
0/0.00
|
|
View(er)s
|
|
2/80
|
|
|
|
|
|
|
| Prev |
Next |
|