Title :
The fast stroke filter
Author :
Su, Yi-Ching ; Lee, Chen-Wei ; Yang, Yung-Hsuan
Author_Institution :
Dept. of Inf. Eng., I-SHOU Univ., Kaohsiung, Taiwan
Abstract :
A stroke is defined as a straight line or arc corresponding to a segment of character, and a character is composed of one or several strokes. A novel fast stroke filter which can localize stroke-like structures with very high speed has been proposed in this paper. Compared with related work, our fast stroke filter attacks the stroke detection problem in a more efficient way. It uses line segments to sampling the intercept of stroke body along both perpendicular axes, and then stroke orientation can simply be determined by a set of qualification rules. Much computing redundancy existing in the original stroke filter can be reduced, and the computation scale has been downsized from rectangle to line. According to our experimental results, fast stroke filter is superior to original stroke filter in both speed and quality, and the averaged speed-up is over 30 times.
Keywords :
character recognition; image recognition; text analysis; character segment; fast stroke filter; qualification rules; stroke like structures; Classification algorithms; Filtering algorithms; Filtering theory; Matched filters; Pixel; Qualifications; License plate detection; Stroke filter; Stroke structure; Text detection;
Conference_Titel :
Machine Learning and Cybernetics (ICMLC), 2010 International Conference on
Conference_Location :
Qingdao
Print_ISBN :
978-1-4244-6526-2
DOI :
10.1109/ICMLC.2010.5580744