Context Inference in Region-Based Image Retrieval

In this paper, a method for inference of high-level semantic information for image annotation and retrieval is proposed. Bayesian theory is used as a tool to model a belief network to configure semantic labels for image regions. These semantic labels for regions are obtained from a multi visual feature-based object detection approach. The aim is to model potential semantic descriptions of basic objects in the images, the dependencies between them, and the conditional probabilities involved in those dependencies. This information is then used to calculate the probabilities of the effects that those objects have on each other in order to obtain more precise and meaningful semantic labels for the whole images. However, the proposed method is not restricted to the specific region-based approach used in this paper. Rather, the proposed method can be applied in any region-based image retrieval systems. Selected experimental results are presented to show the improved retrieval performance of the proposed method.