Conceptual description of visual scenes from linguistic models

As model-based vision moves towards handling imprecise descriptions like a long bench is in front of the tree, it has to confront questions involving widely variable shapes in unclear positions. Such descriptions may be said to be "conceptual" in the sense that they provide a loose set of constraint s permitting a range of instantiations for the scene. One of the validation s of a computational system's ability to handle such descriptions is provid ed by immediate visualization, which tells the user whether the bench is of the right shape and has been positioned correctly. Such a visualization mu st handle impreciseness in Shape and Spatial Pose, and, for dynamic vision. Object Articulation and Motion Parameters as well. The visualization task is a concretization which consists of generating an "instance" of the scene /action being described. The principal requirement for concretizing the con ceptual model is a large visual database of objects and actions, along with a set of constraints corresponding to default dependencies in the domain. In our work, the resulting set of constraints is combined using multidimens ional fuzzy functions called continuum fields (potentials). A set of experi ments was conducted to determine the parameters of these continuum fields. An instance is generated by identifying minima in the continuum fields invo lved is generated by identifying minima in the continuum fields involved in generating the shape, position and motion. These are then used to create d efault instantiations of the objects described. The resulting image/animati on may be considered to be the "most likely" visualization, and if this mat ches the linguistic description, the continuum fields selected are a good m odel for the conceptual content in the linguistic model of the scene. We pr esent examples of scene reconstruction from conceptual descriptions of urba n parks. (C) 2000 Elsevier Science B.V. All rights reserved.