Enhancing Semantic Features with Compositional Analysis for Scene Recognition Miriam Redi and Bernard Merialdo EURECOM, Sophia Antipolis 2229 Route de Cretes Sophia Antipolis {redi,merialdo}@eurecom.fr Abstract. Scene recognition systems are generally based on features that represent the image semantics by modeling the content depicted in a given image. In this paper we propose a framework for scene recognition that goes beyond the mere visual content analysis by exploiting a new cue for categorization: the image composition, namely its photographic style and layout. We extract information about the image composition by storing the values of affective, aesthetic and artistic features in a compo- sitional vector. We verify the discriminative power of our compositional vector for scene categorization by using it for the classification of images from various, diverse, large scale scene understanding datasets. We then combine the compositional features with traditional semantic features in a complete scene recognition framework. Results show that, due to the complementarity of compositional and semantic features, scene cat- egorization systems indeed benefit from the incorporation of descriptors representing the image photographic layout (+ 13-15% over semantic- only categorization). 1 Introduction The automatic recognition of visual scenes is a typical, non-trivial computer vi- sion task. The aim is to automatically identify the place where a given image has been captured, or, for example, the type of environment in which a robot is navigating. The general approach is to build a statistical model that can distin- guish between pre-defined image classes given a low-dimensional description of the image input, namely a feature vector (here also signature or descriptor ). Fig. 1. Similar images share similar compositional attributes: depth of field for mon- uments, point of view for sports field, contrast for natural scenes, level of details and order for indoor scenes. A. Fusiello et al. (Eds.): ECCV 2012 Ws/Demos, Part III, LNCS 7585, pp. 446–455, 2012. c  Springer-Verlag Berlin Heidelberg 2012