Scene classification using color and structure-based features

Study of scene understanding is a significant challenge. Many conventional methods proposed by these studies have been used or applied for many fields, for instance, scene recognition system for digital camera, similar image retrieval system on websites, and robot vision for autonomous or assist robots. From above, scene understanding is important, however it is as difficult as generic object recognition due to the diversity of categories. Many conventional methods have been proposed, and these focus on color or spatial frequency features in images. Especially, scene classification using features of spatial frequency show efficacy. Seen from the results of these studies, it seems that there is common features within a same scene. In this paper we proposed scene classification method with a focus on the structure of scene. We define the structure of scene as a set of lines in images and calculate these features using Hough space acquired by applying Hough transform to images. In addition, we calculate color features and combine those features. By using these two features we generate two strong classifiers with Boosting algorithm, and combine the results of each strong classifier. To test our approach, we executed two classes classification of scenes for each category using scene classification dataset. The results show that our approach is effective for several scenes especially the scene with artifacts.