Japanese / English

Detail of Publication

Text Language Japanese
Authors Nobuhiko Asada,Masakazu Iwamura,Koichi Kise
Title Improvement of Word Recognition Accuracy with Spellchecker Based on Tendency of Recognition Error of Characters
Journal IEICE Technical Report
Presentation number PRMU2010-268
Location 茨城県つくば市
Reviewed or not Not reviewed
Presentation type Oral
Month & Year March 2011
Abstract In this paper, we discuss a method to cope with severe perspective distortion to realize \all-round recog- nition" of characters and words in the environment. We have proposed a word recognition method satisfying three requirements: ready for real-time processing, free from layout constraints and robust to perspective distortion. The method has the capability to recognize characters captured from the angle of 45 deg. However, in order to realize the all-round recognition, further robustness is required. Thus in this paper, we introduce the GNU Aspell to rectify the recognition results based on the tendency of recognition errors. For the sake of that, we introduce Metashape based on the recognition errors instead of Metaphone of the Aspell based on phonemes. In the experiments, we found that the proposed method improves the recognition rates for words captured from the angle of 20 deg. from about 24% to about 74% at largest.
Back to list