Spatial envelope and background knowledge for scene classification problem

Scene classification problem is one of the major fields of research in artificial vision. The ability to assign the correct label to a scene can provide a significant advantage to automatic processes in order to achieve their task. This paper explores the possibility to classify a scene using objects as attributes and a discrete spatial envelope theory. The challenge is to be able to distinguish among all the existing objects the most discriminative ones in the scene using a proposed background knowledge and sorting functions. The classification process is then guided by a proposed discrete spatial envelope theory in order to provide an accurate and coherent category of scene. The proposed approach offers very satisfying results going up to 69.92% of well classified scenes on the very challenging SUN397 dataset. Compared to some existing state of the art methods, the proposed approach distinguishes itself by proposing a higher rate of classification.