Volume 7, Issue 2 (6-2015)                   IJICTR 2015, 7(2): 59-70 | Back to browse issues page

XML Print


Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:

Keyvanpour M, Tavoli R, Mozaffari S. HWS: A Hierarchical Word Spotting Method for Farsi Printed Words Through Word Shape Coding. IJICTR. 2015; 7 (2) :59-70
URL: http://ijict.itrc.ac.ir/article-1-102-en.html
1- Department of Computer engineering, Alzahra University Tehran, Iran
2- Departments of Computer Engineering, Islamic Azad University, Qazvin Branch Qazvin, Iran
Abstract:   (1160 Views)
Word shape coding (WSC) is a method of document image retrieval (DIR) based on keyword spotting. By using this method, a word can be recognized in the document image, only by identifying some of the features of the word. In this paper, a hierarchical word spotting method, namely HWS, is presented for Farsi document image retrieval through WSC. In HWS method, document images are retrieved by using a new indexing method. In HWS, at first the words in the document images are shape coded based on topological properties. These features include number of sub-words, ascenders, descenders, and holes.A new feature that has been used for this paper is dot's position in word. Six features are obtained which are one top dot, two top dots, three top dots and one bottom dot, two bottom dots, and three bottom dots. Precision of retrieval increases by using these features. Then, all of the shape codes are indexed by building a tree. Retrieval is done based on keyword query in the tree. The results show that the proposed technique is very fast for large volumes of documents. Time complexity for successful and non-successful searching is O(logkn) .This value is better than values in ordinal method. Also, time complexity for indexing is O(logkn) . The HWS method is tested on Bijankhan database. 87867 common words from this database are used for building the dictionary. Test results show that average of precision is 0.83 and average recall is 0.94.
Full-Text [PDF 1015 kb]   (469 Downloads)    
Type of Study: Research | Subject: IT

Add your comments about this article : Your username or Email:
CAPTCHA code