Some variants of Hausdorff distance for word matching

Andrey Andreev, Nikolay Kirov

Several recently proposed modifications of Hausdorff distance (HD) are examined with respect to word image matching for bad quality typewritten Bulgarian text. The main idea of these approaches presumes that omission of the extreme distances between the points of the compared images eliminates the noise (to some extent) and the algorithms become more robust. A few robust HD measures, namely, censored HD, LTS-HD, and a new binary image comparison method that uses a windowed Hausdorff distance, lie in the base of the computer experiments carried out using 54 pages of typewritten text.