Abstract
This paper presents a novel approach using a context-sensitive Bayesian network for natural scene modeling and classification. In contrast to the common approach using of semantic features, we learn the major spatial arrangement (spatial and context information) of scenes and relationships between local semantic concepts and global scene meanings using a contextual Bayesian network. Images' scene probabilities are inferred in a two-level process based on characteristic objects in the image as well as spatial arrangements of key entities through the Bayesian network. We demonstrate the promise of this Bayesian network approach on a set of natural scenes, comparing it with existing state of art approaches.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have