Research of Image Scene Classification Algorithm Based on the Contextual Semantic Information

The conventional Bag of Visual Words model (BoVW) only uses the image feature domain and neglects the contextual semantic information of the image. This paper proposes an image scene classification algorithm based on the contextual semantic information. Firstly, based on the conventional BoVw, we introduce the Markov Random Field model to modeling the image contextual semantic information. Secondly, we use a density-based adaptive selection method to choose the number of the of the optimal probabilistic latent semantic analysis model. At last, we use the Support Vector Machine (SVM) to perform scene classification. The experimental results show that based on the conventional Bag of Visual words model, our algorithm can effectively utilize the contextual semantic information of images and improve image scene classification performance and enhance the classification accuracy.