Research on HVS-Inspired, Parallel, and Hierarchical Scene Classification Framework

Abstract

A novel bionic, parallel, and hierarchical scene classification framework is presented in this paper. Moreover, we build the model based on the perception as defined by the human visual system. At first, we use an image pyramid to present both the global scene and local patches containing specific objects. Second, we build our own codebooks, which satisfy both long stare and short saccade similar to humans. Next, we train the visual words by generative and discriminative methods, respectively, which could obtain the initial scene categories based on the potential semantics using the bag-of-words model. Then, we use a neural network to simulate a human decision process. This leads to the final scene category. Experiments show that the parallel, hierarchical image representation, and classification model obtain superior results with respect to accuracy.