Binary Pattern Descriptors for Scene Classification

Abstract

Scene classification is a computer vision task that aims to identify the kind of scene (such as forest, mountain, beach, etc.) where a picture was taken. Scene classification has application in the development of automatic surveillance systems, robotic navigation, content-based image retrieval systems among other areas. According to how a scene is recognized, scene classification algorithms can be divided in two categories: based on object recognition and based on low-level image features obtained by applying descriptors. This paper proposes a new binary descriptor called Local Lineal Binary Pattern and a new framework that allows the combination of the new binary descriptor with another local and global descriptors in order to improve the automatic classification of scene images. This new binary descriptor increases the spatial support for its calculation allowing to add more spatial information than the traditional binary descriptors such as the Local Binary Pattern and the Modified Census Transform. Experiments conducted over indoor and outdoor scene datasets show that the new proposed descriptor and framework help to improve the results obtained by related works.