Image semantic recognition scheme with semantic-binding hierarchical visual vocabulary model

With the research of interest point detecting and local feature extracting, local feature has been proved to be an effective feature in image/video analysis. And the BOVW (bag of visual words) model aims to cluster local features and generates the visual vocabulary in order to recognize the image semantic. However, the traditional BOVW model has two major drawbacks: 1) semantic loss; 2) no efficient structure of vocabulary to recognize multi-level semantic. In this paper, a semantic-binding hierarchical model is proposed and then the hierarchical visual vocabulary is introduced based on the novel semantic-binding hierarchical model. We propose a scheme where the visual vocabulary is constructed and apply it in multimedia content semantic analysis. The experiment results show that our scheme is effective and efficient compared with some related works on the dataset of Labelme project from MIT.