Grouping For Simplicity

This paper provides an information theoretic approach that reinterprets the grouping process as a model selection process. MDL approximation is used to show that increasingly economical structural models (groups) are selected while combining lower level primitives to form higher level structures. The central idea is that the description length of a 'whole' is better than that of its 'parts'. The notion is applied to form structural interpretations of images under partial occlusion, prior to the application of domain knowledge. From initially fitted segments progressive groups are formed leading to closed structures that are eventually grouped into foreground and background.