Classifier fusion for post-classification of textured images

In this paper, we present various approaches for combining classifiers to improve classification of textured images, which are not generally used in this application framework. This is what we call post-classification step of textured images. Three approaches to combine classifiers are presented: the majority voting approach, belief approach, and classification-based approach. Belief, majority voting and classification-based approaches are compared for classification of real world-data that are sonar images. The obtained results show the interest of this post-classification step, particularly with the belief approach, to improve textured image classification results.