![]() |
![]() |
![]() |
@inproceedings{DBLP:conf/cikm/Kamel96, author = {Ibrahim Kamel}, title = {Fast Retrieval of Cursive Handwriting}, booktitle = {CIKM '96, Proceedings of the Fifth International Conference on Information and Knowledge Management, November 12 - 16, 1996, Rockville, Maryland, USA}, publisher = {ACM}, year = {1996}, pages = {91-98}, ee = {db/conf/cikm/Kamel96.html, http://doi.acm.org/10.1145/238355.238447}, crossref = {DBLP:conf/cikm/96}, bibsource = {DBLP, http://dblp.uni-trier.de} }BibTeX
This paper presents an indexing method that can be used to search a large collection of cursive handwriting and which reports all strings that are similar to a query string. The basic idea is to segment each cursive string into a set of strokes. Each of these strokes can be described with a set of features and, thus, can be stored as points in the feature space. Subsequently, any multi-dimensional access method, such as the R-tree, can be used. Similarity search can be performed by executing a few range queries and then applying a simple voting algorithm to the output to select the strings that are most similar to the query. The proposed index can support similarity queries as well as substring matching. It is resilient to the kind of errors that result from the segmentation process, namely, stroke insertion/deletion and m-n substitution.
We implemented the proposed indexing method and compared it with the only known algorithm (VUE) for searching cursive handwriting. Our proposed index achieves substantial saving in search time over the VUE algorithm and improves the matching rate up to 46% over the VUE algorithm. We also use the Karhunen-Loeve transform to reduce the number of features (data dimensionality) and thus the index size. Our experiments have shown that the extra effort we spent in mapping the data to lower dimensionality space pays off. With a sacrifice of less than 10% of the matching accuracy we reduced the index size by about 45%.
Copyright © 1996 by the ACM, Inc., used by permission. Permission to make digital or hard copies is granted provided that copies are not made or distributed for profit or direct commercial advantage, and that copies show this notice on the first page or initial screen of a display along with the full citation.