Psychophysical analyses of contour processing in humans: the case for qualitative tests

摘要

In this paper, we present the case for qualitative analyses of human vision to provide constraints on computational models. The case is illustrated by examples of experiments that assess: (1) grouping by collinearity across corner junctions (experiment 1) and (2) the role of edges and contrast polarity in grouping (experiment 2). The experiments suggest that grouping by collinearity does operate across corners (though in a weakened form) and that grouping by collinearity but not by surface features is insensitive to the contrast polarity of elements. We outline how such qualitative tests of grouping constrain the development of computational architectures.