A Bayesian network-based framework for semantic image understanding

摘要: Current research in content-based semantic image understanding is largely confined to exemplar-based approaches built on low-level feature extraction and classification. The ability extract both features perform knowledge integration of different types expected raise a new level. Belief networks, or Bayesian networks (BN), have proven be an effective representation inference engine artificial intelligence expert systems research. Their effectiveness due the explicitly integrate domain network structure reduce joint probability distribution conditional independence relationships. In this paper, we present general-purpose framework that employs BN integrating features. efficacy demonstrated via three applications involving pictorial images. first application aims at detecting main photographic subjects image, second selecting most appealing event, third classifying images into indoor outdoor scenes. With these diverse examples, demonstrate engines can within powerful flexible according specific available training data solve inherently uncertain vision problems.