A multiple region selection based approach for scene recognition

Scene recognition is a frequently-studied topic of computer vision. In this work, we propose a solution to this problem that involves multiple-region selection and multiple instance classification. In the proposed approach, first, meaningful and discriminative sub-regions are extracted and then, information coming from these regions are considered within a Multiple Instance Learning framework. The results obtained via the tests performed on 15-Scenes benchmark dataset show that the proposed approach is promising for the classification performance.