Object categorization using Cartesian Genetic Programming (CGP) and CGP-Evolved Artificial Neural Network

In this paper, we address the problem of recognizing object categories by proposing a learning model based on evolutionary algorithm that takes unsegmented, complex images which is tolerant to 2D affine transformations such as scaling and translation in the image plane and 3D transformations of an object such as illumination changes and rotation in depth. To achieve this, first object features are extracted from an image using modified Bag of Keypoints model and then learning and classification is performed through evolutionary network classifiers i.e. Cartesian Genetic Programming (CGP) and Cartesian Genetic Programming Evolved Artificial Neural Network (CG-PANN). Our empirical evaluations show that proposed network classifiers exhibit outstanding ability of learning from fewer training examples with good accuracy. Results are compared with NEAT-Evolved Artificial Neural Network classifier which shows clearly that our network classifiers outperform and generalize better than NEAT.