Automatic scene understanding and labeling

Our main goal is to build a system capable of automatically forming a semantic, easily understandable, description of an image. We will design, extend, and test algorithms, and build software that can enable automated annotation, browsing, and context based search and retrieval from a huge database of image and video data. During the phase II, we will continue to build on the phase I image understanding framework, while extending the technology to include video sources, incorporating new feature-types and learning schemes to improve robustness of labeling.