Why are our similarities so different A reply to Humphreys and Riddoch

Abstract

Humphreys and Riddoch (2002: hereafter, H&R) entitle their commentary on Laws and Gale (2002) with the following question: “Do pixel level analyses describe psychological perceptual similarity”? (We will use the term perceptual similarity to refer also to structural similarity as used by Humphreys and colleagues). This perhaps betrays some misunderstanding of our intention. Throughout our paper, we specifically refer to visual overlap – which we take to mean retinotopic similarity at the pixel level. This differs from notions of psychological or perceptual similarity and certainly differs from the position advanced by Humphreys and colleagues. Although we do not view the two approaches as mutually exclusive, we have some reservations about the utility of perceptual similarity as measured by Contour Overlap (CO) and partonomic features (Humphreys et al., 1988). Moreover, on grounds of parsimony, it is important to examine the role of low-level variables in object recognition and category specificity before turning to high-level variables i.e. psychological/perceptual variables. In response to the commentary by H&R (2002), we would like to draw attention to some issues that relate to: (a) the points they raise about Euclidean Overlap (EO); and (b) difficulties with their conception of perceptual similarity.