A New Background Feature for DHMMs-Based Character Recognition
This paper presents a new recognition scheme using the background feature and Discrete Hidden Markov Models (DHMMs) for off-line handwritten character recognition.The proposed feature extraction method is based on concavity information which is an improvement of Alceus method with 4 directions. In the improved background feature extraction process, each background pixel is scanned from 8 directions.Then if at least two consecutive directions find black pixels, a label is assigned to the background pixel using concavity configurations. Finally the number of background pixels that belong to a specific concavity configuration consists of a feature vector. The Experiments on off-line handwritten Chinese amount in words with different HMMs topologies show that the proposed method is superior to Alceus method. Moreover, different states number is also tested from 5 to 14 to find the better one for the new approach.
Feature Extraction Background Feature Extraction Discrete Hidden Markov Models Off-Line Handwritten Character Recognition
WANG Xianmei FENG Jun YANG Yang LIN Ziyu
Department of Electronics and Information Engineering University of Science and Technology Beijing H Department of Computer Science Shijiazhuang Railway Institute,Shijiazhuang, Hebei Province 050043, C Department of Communication Engineering University of Science and Technology Beijing Haidian, Beijin Zhonghuan Metallurgical Corporation No.56 Andingmenwai Street, Dongcheng district Dongcheng, Beijing
国际会议
厦门
英文
811-815
2006-07-27（万方平台首次上网日期，不代表论文的发表时间）