:: Abstract


 PatternsPossibility-Based Recognition:  A Reflection on Recognition of Shap


Prof. Kambiz Badie


Both "probability" and "possibility" have long been considered as key items for managing uncertainty in problem-solving / decision-making issues like classifying different types of patterns. While the major concern of probability is to emphasize on the observations from experiencing different cases, possibility mostly emphasizes on the aspect of structurality with regard to a situation. In this way, the necessary condition for justifying the utility of possibility for managing uncertainty within the process of pattern classification is the ability of considering structurality for the ongoing patterns. It is to be noted that, the structurality of a pattern or a situation goes back to the fact that how its parts or components can comply with certain models with certain characteristics. In this way, "possibility-based recognition" is concerned with a range of possibilities with regard to a variety of significant substructures in a pattern, which come together to give sense/identity to that pattern or situation on the base of the models belonging to these substructures. It is believed that such an approach to recognition takes place on the base of a kind of information fusion whose function is to incorporate a number of possibilities regarding the existing substructures in a pattern in order to minimize plausibly the very uncertainty which exists within the process of pattern classification. The role of fusion in our approach is that, once the vector representing a pattern was constructed on the base of the information on average possibility degrees for a variety of significant substructures in that pattern, it is first tried to determine the value of similarity between the components of this vector and the corresponding significant substructures in the existing classes taking into account a kind of similarity function and then determine the products of these similarity values for each class. Obviously, the class for which the value of this product is the most, should be considered as the most suitable class for the corresponding pattern. Fusion of possibility degrees can be performed at two levels; (i) characterizing the significant substructures in a pattern with the purpose of determining the most reasonable class for it, and (ii) final classification of a pattern including multiple classes.


As examples of patterns with structural nature, we may mention "patterns of shapes" consisting of a variety of curve segments or curve/region boundaries, "patterns of semantic entities" containing lexicons with certain meanings, and "patterns of mathematical/logical expressions" containing logical symbols, predicates, arguments, functions and variables. What we try to clarify in this presentation is how "possibility-based recognition" in the way explained above has the ability to perform a plausible classification of "patterns of shapes" taking into account the role of information fusion. Within the scope of shapes, some examples are discussed regarding classification of both Latin and Arabic characters, mentioning how based on the possibility degrees belonging to the corresponding substructures in these characters, their classification can be performed in a plausible manner. Shapes in general include "objects" (living, non-living and artificial), phenomena (cosmic, social, organizational and personal) and symbols (alpha-numeric and conventional configurations including message).


As a conclusion, it is significant to notice that the possibilistic approach to recognizing structurally-rich patterns provides a conducive opportunity to consider only those significant substructures, which are peculiar to the identity of the classes, thus avoiding the information regarding the irrelevant substructures that makes the process of classification rather time-consuming. Interestingly, the process of intuition in human being, which is a sort of high-level perception, seems to approve this fact.

Key Words—Perception, possibility, uncertainty, pattern classification, fusion, substructures, shape.        



IEEE Index


Submit Your Paper 

Users Login             

Reviewers Login     

Paper Templates


Video news