IMAGE-BASED OBJECT RECOGNITION IN MAN, MONKEY AND MACHINE

Theories of visual object recognition must solve the problem of recogn izing 3D objects given that perceivers only receive 2D patterns of lig ht on their retinae. Recent findings from human psychophysics, neuroph ysiology and machine vision provide converging evidence for 'image-bas ed' models in which objects are represented as collections of viewpoin t-specific local features. This approach is contrasted with 'structura l-description' models in which objects are represented as configuratio ns of 3D volumes or parts. We then review recent behavioral results th at address the biological plausibility of both approaches, as well as some of their computational advantages and limitations. We conclude th at, although the image-based approach holds great promise, it has pote ntial pitfalls that may be best overcome by including structural infor mation. Thus, the most viable model of object recognition may be one t hat incorporates the most appealing aspects of both image-based and st ructural-description theories. (C) 1998 Elsevier Science B.V. All righ ts reserved.