Ensemble of Two-Class Classifiers for Image Annotation

Image annotation can be formulated as a multi-class classification problem. A multi-class classification problem can be solved by ensemble classifiers. We investigate the ensemble of multiple two-class classifiers based on MPEG-7 standard. To get ride of redundancy information, a binary-coded chromosome genetic algorithm is used to select individual optimal classification feature pattern for each two-class other than for all the classes. Two-class classifiers are generated based on the results of feature selection, and majority voting scheme is used to combine two-class classifiers. The experimental results over 2000 classified Corel images show that our approaches can improve annotation accuracies.