| | |
Summary: Online Time Stamped Text Indexing
Amihood Amir y Gad M. Landau z Esko Ukkonen x
Bar-Ilan University Haifa University University of Helsinki
and and
Georgia Tech Polytechnic University
Abstract
In this paper we present an eÆcient method to index a text stream on-line in a fashion that
allows, at any point in time, to nd the longest suÆx of the text that has appeared previously,
and the closest (farthest) times in which it has appeared. Our algorithms allow eÆcient answers
to these queries.
1 Introduction
Many pattern recognition tasks are solved by ad-hoc heuristics that try to exploit special knowledge
of domain properties with varying degrees of success. Recently, there have been attempts to solve a
class of pattern recognition problems by analysis of repetitions and periods in a sequence of images
taken over time [10]. These methods are more general and not domain-speci c.
Johansen [10] used this idea for surveillance, character labeling, and discrimination of handwriting
and texture. In the surveillance application, the goal is nding \unexpected" changes in a sequence
of photographs, when there is no prior de nition of \unexpected". The \surprising" nature of an
image is to be detected from the syntax of the sequence of images. Johansen uses exact matching
in his applications.
|
