Perceptual categories for spatial layout

Abstract

The central problems of vision are often divided into object identification and localization. Object identification, at least at fine levels of discrimination, may require the application of top–down knowledge to resolve ambiguous image information. Utilizing top–down knowledge, however, may require the initial rapid access of abstract object categories based on low–level image cues. Does object localization require a different set of operating principles than object identification or is category determination also part of the perception of depth and spatial layout? Three–dimensional graphics movies of objects and their cast shadows are used to argue that identifying perceptual categories is important for determining the relative depths of objects. Processes that can identify the causal class (e.g. the kind of material) that generates the image data can provide information to determine the spatial relationships between surfaces. Changes in the blurriness of an edge may be characteristically associated with shadows caused by relative motion between two surfaces. The early identification of abstract events such as moving object/shadow pairs may also be important for depth from shadows. Knowledge of how correlated motion in the image relates to an object and its shadow may provide a reliable cue to access such event categories.

Diagnostic Colors Mediate Scene Recognition

“…The notion of a scene context, scene ''gist,'' or scene spatial layout, however, remains to be precisely defined (Rensink, 2000). In fact, it is only recently that its role in the recognition of complex pictures has been acknowledged (Epstein & Kanwisher, 1998;Kersten, 1997;Schyns & Oliva, 1994;Sanocki & Epstein, 1997). One goal of this study was to further specify the information content of the scene gist and to demonstrate its sufficiency for scene categorization.…”

The knowledge used in vision and where it comes from

“…In this volume these ideas are particularly important for the papers by and Hinton & Ghahramani (1997), and some of the nicest examples of perceptual e¡ects that may result from redundancy analysing mechanisms are to be found in the paper by Kersten (1997). The prominent use of the KarhunenL oe¨ve transform (KLT) or principal components analysis (PCA) in the papers by Edelman (1997), Bobick (1997), Taylor (1997) and Pentland (1997) show that the bene¢ts of analysing covariance are recognized in these areas of computer vision.…”

“…Judgements of size and distance provide many examples of illusions that might result from repulsion induced by past experience of associations, and some of these include: (i) impression of distance from haze and blue tinge; (ii) micropsia induced with a prism or concave lens; (iii) the`toy-town' illusion from an excess range of disparities; (iv) illusions of depth and slant from vertical disparities; (v) false rotation of a`concave' face; (vi) many e¡ects described by Kersten (1997). One should always consider contingent adaptation before accepting innate mechanisms or top-down in£uences.…”

A Limit to the Speed of Processing in Ultra-Rapid Visual Categorization of Novel Natural Scenes

“…The nature of the underlying mechanisms is currently the subject of intense debate (Wallis & Bu Èlthoff, 1999;Treisman & Kanwisher, 1998;Biederman & Kalocsai, 1997;Tanaka, 1997;Logothetis & Sheinberg, 1996), but it is generally believed that this sort of object recognition must involve bidirectional information processing, in which bottom-up information originating in the retina interacts with feedback mechanisms in a system preset by top-down knowledge (Rao & Ballard, 1999;Sastry, Shah, Singh, & Unnikrishnan, 1999;Mesulam, 1998;Przybyszewski, 1998;Vecera & O'Reilly, 1998;Frith & Dolan, 1997;Grossberg, Mingolla, & Ross, 1997;Humphreys, Riddoch, & Price, 1997;Kersten, 1997;Tallon-Baudry, Bertrand, Delpuech, & Permier, 1997;Ullman, 1996;Kosslyn, 1994). Indeed, recent neurophysiological studies have demonstrated the importance of lateral and feedback interactions in fundamental visual processes that include texture segmentation and figure± ground effects (Hupe Âet al, 1998;Lamme, Super, & Spekreijse, 1998) as well as visual attention (Desimone, 1998;Hillyard, Teder-Salejarvi, & Munte, 1998;Roelfsema, Lamme, & Spekreijse, 1998;Vidyasagar, 1998;Duncan, Humphreys, & Ward, 1997;Luck, Chelazzi, Hillyard, & Desimone, 1997).…”