![]() | Only 14 pages are availabe for public view |
Abstract The statistical approach to pattern recognition is among the early approaches applied in this Field. Statistical classifiers are devided into two main categories, the parametric classifiers and the nonparametric classifiers. The main drawback of the parametric type is that it gives weak results on unknown data distributions. On the other hand, the nonparametric methods require extensive amounts of design samples, storage capacity and computing power in order to give good approximations of our data distributions. In this thesis a semiparametric model consisting of a mixture of component densities, whose Parameters are optimized using a hybrid genetic algorithm is proposed. The optimized Density functions using this model can be quite general, they are not directly parammetrized Although their components are. In our work we tried to make cooperation between a speedy local search gradient algorithm, The Gaussian clustering algorithm and the more global perspective of a genetic algorithm. This proposed technique was tested by applying it to a number of classification problems Ranging from small standard data to large scale data set of fingerprinting classification and the experimental results show that the achieved improvement of classification error compared to many classifiers representing the last state of the art is significant. |