Font Size: a A A

A Uighur New Writing Recognition Tecnoogy Based On Feature Matching

Posted on:2015-12-24Degree:MasterType:Thesis
Country:ChinaCandidate:F ZhaoFull Text:PDF
GTID:2298330431492102Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
In order to electronic paper materials of the Uighur New Writing, the paper is devoted tostudy the recognition technology of printed Uighur New Writing document. Recognition ofUighur New Writing use a feature-based template matching method. Some feature dimensionsare large and others are small, the dimensions with large in recognition is generally moretime-consuming but high identification accuracy, generally, it’s more time-consuming but canget a high identification accuracy to recognize latter by a large-dimension feature. Similarly,it’s less time-consuming but can get a low accuracy when using a small-dimension feature.But the ideal recognition must have the high speed and high accuracy. To solve this problem,a two-stage character matching recognizer is designed. The principle is that the first stagerecognizer uses the Character feature of small dimensions to position a small template libraryin the library of character feature template. In the second stage recognizer, we use thecharacter feature of large dimensions to accurately identify letters in small template library.The test showed that this recognizer has both fast speed of the low dimensions features thehigh accuracy of high dimensions feature. The amount of same shape letters is about twofifths of the amount of all letters, so, after the simple recognition, it will find that the sametype letters recognition error phenomenon is widespread and this seriously impact on theoverall recognition rate. To solve this problem, the paper designed a method to distinguish thesame shape letters, mainly based that the outline size of uppercase letters is bigger than theone of the lowercase letters, This method is simple and easy to use, the test showed that it canwell distinguish these same type letters. Tested by the experiment, when the template librarysize is10574, the recognition rate of the system can reach about98%.
Keywords/Search Tags:Uighur New Writing, Feature matching, Feature Extraction, The same type ofletter, Character Recognition
PDF Full Text Request
Related items