Hausdorff distances for searching in binary text images

Abstract

Hausdorff distance (HD) seems the most efficient instrument
for measuring how far two compact non-empty subsets of a metric space are from each other. This paper considers the possibilities provided by HD and some of its modifications used recently by many authors for resemblance between binary text images. Summarizing part of the existing word image matching methods, relied on HD, we investigate a new similar parameterized method which contains almost all of them as particular cases. Numerical experiments for searching words in binary text images are carried out with
333 pages of old Bulgarian typewritten text, 200 printed pages of Bulgarian Chrestomathy from year 1884, and 200 handwritten pages of Slavonic manuscript from year 1574. They outline how the parameters must be set in order
to use the advantages of the proposed method for the purposes of word matching in scanned document images.