Handling Uncertainty in a System for Text-Symbol Context Analysis

In pattern analysis, information regarding an object can often be drawn from its surround­ ings. This paper presents a method for handling uncertainty when using context of symbols and texts for analyzing technical drawings. The method is based on Dempster-Shafer theory and possibility theory. 1 Introduction In several pattern analysis problems it is interesting to look at the context in which an object is placed to gain information on what the object is. This article focuses on a method for analyzing one such problem, the problem of combining text and symbols in a system for processing technical drawings. In particular we look at the method for handling uncertainty in this system. In our system we have a pattern recognition program that proposes alternatives for e\·ery text string and every symbol. To all strings or symbols the recognizer gives a probability distribution on the possible alternatives. Our problem is to combine these probabilities with information from the context of the strings and symbols. This problem is essentially the same as combining uncertainty measures from different sources of evidence. One promising candidate for solving this kind of problems is Dempster-Shafer theory [5J, and we shall use this theory to combine our probabilities with the uncertainty measure obtained from the context analysis. This uncertainty measure will be set equal to the plausib ility measure of Dempster-Shafer theory. Results from (4] are used in the argumentation for the ideas behind our method. In the next section a short review of Dempster-Shafer theory is given, and we show sh ow how to combine the probabilities and plausibilities by using a result from [.S]. In section :3 we cliscn.s;; how the plausibilities are obta. ined and give an example of the use of this met ho d. • This paper is a revised version of the paper with the same title presented at the SCAI '88 conference in Tromso. Norway.