Abstract
Image classification represents an important task in machine learning and computer vision. To capture features covering a diversity of different objects, it has been observed that a sufficient number of learning instances are required to efficiently estimate the models' parameter values. In this paper, we propose a genetic programming (GP) based method for the problem of binary image classification that uses a single instance per class to evolve a classifier. The method uses local binary patterns (LBP) as an image descriptor, support vector machine (SVM) as a classifier, and a one-way analysis of variance (ANOVA) as an analyser. Furthermore, a multi-objective fitness function is designed to detect distinct and informative regions of the images, and measure the goodness of the wrapped classifiers. The performance of the proposed method has been evaluated on six data sets and compared to the performances of both GP based (Two-tier GP and conventional GP) and non-GP (Naïve Bayes, Support Vector Machines and hybrid Naïve Bayes/Decision Trees) methods. The results show that a comparable or significantly better performance has been achieved by the proposed method over all methods on all of the data sets considered.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have