Parts of recognition

It is proposed that, for the task of object recognition, the visual system decomposes shapes into parts using a rule defining part boundaries rather than part shapes, that the rule exploits a uniformity of nature—transversality, and that parts with their descriptions and spatial relations provide a first index into a memory of shapes.

Abstract

We propose that, for the task of object recognition, the visual system decomposes shapes into parts, that it does so using a rule defining part boundaries rather than part shapes, that the rule exploits a uniformity of nature—transversality, and that parts with their descriptions and spatial relations provide a first index into a memory of shapes. This rule allows an explanation of several visual illusions. We stress the role inductive inference in our theory and conclude with a pre´cis of unsolved problems. Les auteurs sugge`rent que le syste`me visuel pour la reconnaissance des objets, de´compose les formes ene´le´ments et qu'il utilise pour cela une re´gle de´finissant les fronte´res de cese´le´ments plutoˆt que leurs formes. Cette re`gle exploite une re´gularite´de la nature: la transversalite´. Lese´le´ments, leurs descriptions et leurs relations spatiales fournissent un premier index dans la me´moire des formes. On peut avec cette re`gle rendre compte de plusieurs illusions visuelles. Les auteurs sur le roˆle de l'infe´rence inductive et concluent en indiquant les proble`mes non re´solus.