Functional ‘Tunnel Vision’ in Pattern Classification Learning

Abstract

Visual pattern recognition has been defined in cognitive science as the ability to relate a perceived object to a learned categorical concept (see Rosch and Lloyd, 1978 Cognition and Categorization Hillsdale, NJ: Lawrence Erlbaum). In contrast, physiologists and clinical practitioners often prefer a low-level functional approach, where visual performance is characterised in terms of acuity, visual field, and contrast sensitivity. From a paradigmatic viewpoint, the former definition can be related to tasks of pattern classification, whereas the latter involves tasks of discrimination. We show that this dichotomy concerning perceptual tasks corresponds to dichotomous internal representations that observers develop in such tasks. For a common set of grey-level patterns (compound Gabor signals), we compared performance in classification and discrimination learning in foveal and extrafoveal vision. Learning speed for classification learning was found to be drastically reduced in extrafoveal relative to foveal vision. Discrimination learning of the same signals remained unaffected by viewing condition. Further analysis of the data in terms of a probabilistic prototype model (Rentschler, Juettner, and Caelli, 1994 Vision Research 34 669 – 687) showed that internal representations acquired in extrafoveal classification learning are characterised by a distinct reduced perceptual dimensionality. The results suggest that internal representations underlying pattern classification and discrimination arise at distinct cortical levels, and that the former are developed within an extremely narrow visual field essentially restricted to the fovea.