chapter  3
22 Pages

Visual-Semantic Context Learning for Image Classification

WithQianni Zhang, Ebroul Izquierdo

This chapter provides a forum for the state-of-the-art research in the emerging field of visual-semantic context analysis and addresses the growing interest in automatic classification of multimedia content. It suggests some ideas on an image classification approach that jointly exploits the visual and semantic context embedded in images. The chapter presents a survey of the state-of-the-art in context exploitation for semantic image understanding, with specific focus on visual analysis based on block regions. It introduces the semantic context modeling (SCL) method for automatic context learning and inferences to boost the classification performance based on general region based image classifcation systems. Semantic context learning is achieved by jointly considering the multi-feature similarities extracted from each block using the MFL module, and exploring the relationships between them. The results of Multi-Feature Based Classification and Context Inference Based Classification experiments show that the classes of an image can be effectively determined using the two proposed approaches: MFL and SCL.