A PARALLEL, HIERARCHICAL SCENE CLASSIFICATION FRAMEWORK

摘要

A novel parallel, hierarchical scene classification framework is presented in this paper. At first, we use an image pyramid to present both the global scene and the local patches containing specific objects. Secondly, we build our own codebooks, which extract both global and local features. Next, we train the visual words by generative and discriminative methods respectively, which could obtain the initial scene categories based on the potential semantic that is reasoned by the bag-of-words model. Then, we use a neural network to obtain the final scene categories. Experiments show that the parallel, hierarchical image representation and classification model could obtain better results.