A study in descriptive representation of pictorial data

摘要

Although much effort has been expended on automatic scene description, especially in the various robot and “hand-eye” projects, these efforts have usually been directed toward description for immediate use, i.e. description of a scene for the purpose of stacking blocks using a manipulator arm, or for allowing a robot to move through an environment. This paper is concerned with a somewhat different type of description in which a scene is described in general terms for an unspecified future use. This type of description has application in (1) advanced robot systems, where the robot, similar to the human, will build up an “encyclopedia” of descriptions for possible use, and (2) in question-answering systems for image data bases in which the descriptions are to represent an “encyclopedic” knowledge of the images. Experiments in nongoal-directed description using human subjects are described—experiments which seek to determine how general descriptions are generated and the nature of such descriptions.