Genetic Programming for Multiclass Texture Classification Using a Small Number of Instances
© Springer International Publishing Switzerland 2014. The task of image classification has been extensively studied due to its importance in a variety of domains such as computer vision and pattern recognition. Generally, the methods developed to perform this task require a large number of instances in order to build effective models. Moreover, the majority of those methods require human intervention to design and extract some good features. In this paper, we propose a Genetic Programming (GP) based method that evolves a program to perform the task of multiclass classification in texture images using only two instances of each class. The proposed method operates directly on raw pixel values, and does not require human intervention to perform feature extraction. The method is tested on two widely used texture data sets, and compared with two GP-based methods that also operate on raw pixel values, and six non-GP methods using three different types of domain-specific features. The results show that the proposed method significantly outperforms the other methods on both data sets.