Outline of a New Feature Space Deformation Approach in Fuzzy Pattern Recognition
Petar B. Petrović · FME Transaction · 2003
Pattern recognition ability is one of the most important features that characterize intelligent behavior of either biological or artificial systems. Mathematical pattern recognition is the way to solve this problem using transparent algorithms that are mostly based on conventional mathematics. In complex systems it shows inadequacy, primary due to the needs for extensive computation and insufficient robustness. Algorithms based on soft computing approach offer a good alternative, giving a room to design effective tools for real-time application, having in mind that relevance (significance) prevails precision in complex systems. In this article is modified and extended subtractive clustering method, which is proven to be effective in real-time applications, when massive pattern sets is processed. The new understanding and new relations that connect parameters of the algorithm with the information underlying the pattern set are established, giving on this way the algorithm ability to be data driven to the maximum extent. Proposed algorithm is verified by a number of experiments and few of them are presented in this article .