Catalogue of Artificial Intelligence Techniques
Author(s): Robert B. Fisher
A computer-vision technique in which interpretation of the image data is aided by some form of model of what is expected to be seen in the image. For example, a part inspection system would use a description (model) of where the key features of the part were and where specific tests should be undertaken. The main reasons for this approach are: (i) without some representation of the object no sensible identification is possible, (ii) the use of a geometric model may allow one to infer the position of the object in the scene and (iii) the model can be used to direct the image processing, thus saving considerable processing time. The model may be implicit in the image analysis program, but more sophisticated approaches use explicitly defined models (allowing substitution of new models). The models may be expressed in many forms, including: geometric models (e.g., sizes and positional relationships between features), relational models (e.g., a graph with arcs representing pairwise relationships) or property models (e.g., a table of the expected properties, like area, colour, etc.). There is a fuzzy dividing line between the use of `model-based' for specific objects or object classes, and the use of the term for more generic features, such as lines. That is, a model of a line, or a curve, is a generally useful model and may be as precise and explicit as a model of a real object. In these cases, one often uses the term `knowledge-based' because these features are more abstract, and their identification usually depends more on knowledge of how they appear in an image.
- Fisher, R.B., From Surfaces to Objects: Computer Vision and Three
Dimensional Scene Analysis
, John Wiley and Sons , Chichester , 1989.