Viewpoint-invariant information in subordinate-level object classification

Often a major reason for having humans in a complex human-machine system is that people can readily classify visual images, even when they are novel. These requirements for visual classification are frequently performed at a subordinate level, rather than at a basic level. The employment of a common term - subordinate level - has obscured the heterogeneity of the perceptual processing required to achieve this level of classification. A taxonomy that specifies the information for three kinds of subordinate-level classifications of objects (excluding faces) is proposed. Subordinates in case 1 are distinguished by a representation, a geon structural description (GSD), specifying a viewpoint-invariant characterization of an object's large parts and the relations among these parts. Subordinates in case 2 are also distinguished by GSDs except that the distinctive GSDs are present at a small scale in a complex object so the location and mapping of the GSDs are contingent on an initial basic-level classification. Expertise for cases 1 and 2 can be easily achieved through specification, often verbal, of the GSDs. Subordinates in case 3, which have been the subject of extensive theorizing with "view- based" template models, require fine metric discriminations. Case 1 and 2 account for the overwhelming majority of shape-based basic-level and subordinate object classifications that people can make quickly and accurately.