Scene Categorization by Introducing Contextual Information to the Visual Words

In this paper, we propose a novel scene categorization method based on contextual visual words. In this method, we extend the traditional 'bags of visual words' model by introducing contextual information from the coarser scale level and neighbor regions to the for local region of interest. The proposed method is evaluated over two scene classification datasets of 6,447 images altogether, with 8 aid 13 scene categories respectively using 10-fold cross-validation. The experimental results show that the proposed method achieves 90.30% and 87.63% recognition success for Dataset 1 and 2 respectively, which significantly outperforms previous methods based on the visual words that represent the local information in a statistical manner. Furthermore, the proposed method also out performs the spatial pyramid matching based scene categorization method, one of the scene categorization methods which achieved the best performance on these two datasets reported in previous literatures.