Classifier ensemble based-on AdaBoost and genetic algorithm for automatic image annotation

Image classification approach is one promising method used for automatic image annotation. In order to improve image annotation accuracy, recent researchers propose to use AdaBoost algorithm for the ensemble of classifiers. But as it is difficult for AdaBoost algorithm to search a large feature space, only fewer features are used for the construction of weak classifiers in ensemble. As a result, it is easy to fall into local optimal. We use all the 25 image low-level features of Multimedia Content Description Interface to descript images. Genetic algorithm is used to decrease the search space by randomly select a subset of features. We construct a multi-class weak classifier for each of the features in the subset and their potential combinations respectively. k-nearest neighbor classifier is used as the base classifier and dasiaone vs. onepsila scheme is chosen to build multi-class classifiers. Lastly, we use AdaBoost. M1 algorithm to generate an ensemble classifier and optimize it combining with genetic algorithm. The results of experiment over 2000 classified Corel images show that the ensemble classifier generated in larger search space has higher annotation accuracy.