Loading...
Please wait, while we are loading the content...
Similar Documents
Learning to Recognize Objects Author ( s ) :
| Content Provider | Semantic Scholar |
|---|---|
| Author | Smith, Linda B. |
| Copyright Year | 2008 |
| Abstract | A theory of object recognition requires a theory of shape. Despite considerable empirical and theoretical research, however, a definition of object shape has proved elusive. Two experiments provide new insights by showing that children's object recognition changes dramatically during the period between 17 and 25 months. During this time, children develop the ability to recognize stylized threedimensional caricatures of known and novel objects. This ability is linked to the number of object names in children's vocabularies, suggesting that category learning may be a driving force behind the developmental changes. The central problem for a theory of object recognition is a theory of shape. Although objects seem to be recognized by their shape (Biederman, 1987; Edelman & Duvdevani-Bar, 1997; Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976), real instances of real categories are, in fact, rarely the exact same shape. For example, rocking chairs, stuffed chairs, and desk chairs are the "same shape" only under some highly abstract description. Thus, a definition of shape is the central business and a primary area of contention in the fields of human and machine object recognition (e.g., Edelman & Duvdevani-Bar, 1997; Hummel, 2000). The experiments reported here provide new insights into this issue by showing that an abstract description of shape one under which rocking chairs and desk chairs are the same develops as a product of very young children's category learning. A COMMON IDEA AMONG COMPETING ACCOUNTS According to Biederman's (1987; Hummel & Biederman, 1992) recognition-by-components (RBC) account, objects are represented in terms of a set of simple geometric components called geons such that common objects are readily recognized given only two to three geons in the proper spatial arrangement. By this account, the reason a variety of different chairs are seen as chair shaped is that they all conform to the same abstract and componential representation. The RBC account makes no explicit claims about the developmental origins of these representations. Edelman and his colleagues (Edelman, 1995; Edelman & Duvdevani-Bar, 1997) offer a competing account that begins with the premise that perceivers store view-dependent images of objects (Tarr, 1995). Category learning creates prototypes for (a perhaps small number of) initial categories by interpolating multiple views of multiple instances. Duvdevani-Bar and Edelman (1999) demonstrated that once a number of such prototypes have formed, they serve as landmarks (basis functions) in the shape space such that new objects are represented in terms of their (weighted) distance to those landmarks. In this way, the prototypes for well-learned categories determine the dimensions of the shape space and the perception of even novel things. Although this account differs in important and fundamental ways from RBC, it is like RBC in that it posits shape representations that are an abstraction over the detailed and specific shapes of real things. Again, a variety of chairs can be seen as chair shaped because they are all highly similar under a particular description of the shape similarity space. Edelman's account also offers a clear-cut developmental hypothesis: Shape similarity should change as a product of early category learning.ion over the detailed and specific shapes of real things. Again, a variety of chairs can be seen as chair shaped because they are all highly similar under a particular description of the shape similarity space. Edelman's account also offers a clear-cut developmental hypothesis: Shape similarity should change as a product of early category learning. THE EXPERIMENTAL QUESTION The experiments that follow examined this developmental hypothesis. They did not, however, examine the precise nature of the underlying representations, the main point of contention in the adult and machine literature. The experiments examined very young children's recognition of three-dimensional caricatures of the shapes of common things, such as those illustrated at the top of Figure 1. Recognition of impoverished and stylized forms implies a sparse and thus categoryencompassing description of shape. When in the course of learning early object categories does such an abstract description of shape emerge? The children participating in the experiments were between the ages of 17 and 25 months. This is a period during which children progress from knowing the names of few objects to knowing the names of several hundred objects (e.g., Bloom, 2000) and a period during which attention to shape in naming tasks increases dramatically (e.g., Smith, 1999). |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.iub.edu/~cogdev/labwork/geons.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
