An overview of the role of context in computer vision

This survey investigates the use of context as a method for limiting the search space and improving the output of computer vision systems. It presents current thoughts on the value of using context in intelligent systems and examines the representations and strategies of computer vision systems that make use of it. We define three types of context--immediate, modeled, and operational--and describe its two main roles in computer vision: selection of algorithms and verification of hypotheses. We present common themes of current context-based computer vision systems, highlight some special ideas, and examine the design, processing, and implementation status of five systems in detail. Thoughts on promising ideas and areas that require further investigation are also presented.