摘要
为了解决单个文字识别器无法进一步提高手写文字分类认识率的瓶颈问题,通过构建统合识别器来进一步提高单文字的认识率.以基于MRF的在线识别器和基于MQDF2的离线识别器两种性能互补识别器为基础,根据单个识别器评价值的不同正规化方法,构建了一个具有更高性能的统合识别器.该识别器在东京农工大学手写体数据库为测试数据,统合识别器的1位认识率达到95.36%.与单个的在线识别和离线识别相比,识别率分别提高了3%和5%.
To break the obstacle of further improving the recognition accuracy by single character recognizer, the combined recognizer is proposed to build to improve the recognition accuracy in the paper. Combining two complementary recognizers, namely MRF online recognizer with MQDF2 offline recognizer, an efficient combined recognizer with higher performance is constructed by different methods of normalization in evaluation value of each single recognizer. The Japanese script database of Tokyo University of Agriculture and Techno|ogy is used as the test data of the combined recognizer, and its recognition accuracy can reach 95.36%. Compared with single online and offline recognizer, the accuracy of the combined recognizer is respectively improved by 3% and 5%.
出处
《华北水利水电学院学报》
2012年第5期53-56,共4页
North China Institute of Water Conservancy and Hydroelectric Power
关键词
MRF在线识别器
MQDF2离线识别
统合识别器
评价值正规化
MRF online recognizer
MQDF2 offline recognizer
combined recognizer
normalization in evaluation, ~vaLue~ ,~o