Abstract
This paper deals with a data presentation model based on fuzzy portraits. The fuzzy portraits are formed by integral characteristics of pattern classes. It is the basis for fuzzy classifier construction. It is determined that further division of some classes of images into clusters increases the quality of pattern recognition algorithm. The main idea of fuzzy clustering for fuzzy portraits creating and problem of adequate fuzzy partition choice is considered. The paper provides the stages of fuzzy production knowledge base construction on the basis of fuzzy portraits. The local validity measure for fuzzy portrait is defined. The problem of identification in chemical and food industries is considered as an application of this approach.
Highlights
Pattern recognition problems very often include fuzzy aspects
Many problems in chemical and food industries can be reduced to this modification of the pattern recognition problem
The fuzzy classifiers are applied to pattern recognition and data analysis problems and each fuzzy inference rule has linguistic interpretation
Summary
According to Lotfi Zadeh, the transition between classes is more gradual than intermittent This statement is confirmed by practice of problem solving. In classical pattern recognition problem statement classes of images have to be strongly divided. An algorithm based on fuzzy portraits of classes of images was proposed [3]. These fuzzy portraits describe all objects belonging to a certain class of images in general. The suggestion is to use these portraits for representing the results of algorithm in the form of fuzzy sets. This approach to a certain degree solves the problem of ambiguous results in the case of intersecting classes. The results of this research were presented at the Pattern Recognition and Image Processing Conference in Belarus, Minsk and partially presented in [4]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have