Multilevel grouping: combining bottom-up and top-down reasoning for object recognition

Presents a multilevel grouping scheme that groups primitive image features based on properties of perceptual organization, and groups higher-level structures to more closely approximate complex objects for recognition. At each level, grouping is performed according to geometric relationships among the component objects. The authors´ scheme represents the grouping criteria by fuzzy sets, so that the degree of membership in a grouping criterion set reflects the degree to which the objects fit the criterion. Examples of grouping functions are provided for both top-down and bottom-up grouping