From line drawings to impressions of 3D objects: developing a model to account for the shapes that people see R Cowie* and R Perrott? zyxwvutsrqponmlkjihgfedcbaZYXWVUTSR The interpretations people attach to line drawings reflect shape- related processes in human vision. Their diver- gences from expectations embodied in related machine vision traditions are summarized, and used to suggest how human vision decomposes the task of interpreta- tion. A model called IO implements this idea. It first identifies geometrically regular, local fragments. initial decisions fix edge orientations, and this information constrains decisions about other properties. Relations between fragments are explored, beginning with weak consistency checks and moving to fuller ones. IO’s output captures multiple distinctive characteristics of human performance, and it suggests steady progress tow ards understanding shape- related visual processes is possible. Keywords: shape, line drawing, Impossible Object, line labelling computational modelling This paper is about the way people come to see particular tridimensional shapes when they look at line drawings. It has two related aims. The first is to suggest how studying the interpretation of line drawings can illuminate shape-related processes in human vision. The second is to describe a system (called IO) which interprets line drawings-in a way that captures distinc- tive features of human performance. At the end we will consider how these aims relate to the general progress of computational vision. The basic reason for studying line drawings is that *School of Psychology, Queen’s University, Belfast BT7 INN, UK; ‘Department of Computer Science, Queen’s University, Belfast BT7 INN, UK Paper received: I July 1992; revised paper received: IO December I992 they are challenging. People’s responses to them often run counter to natural intuitions about vision, and that provides an impetus to explore less obvious hypo- theses. This is particularly true when we look beyond drawings which can represent very simple and regular objects. That point is developed in the next section. Line drawings have a specially strong claim on our attention if we are interested in shape, for two reasons. First, the way people see them is likely to be determined by the visual system’s tacit expectations about the types of shape that occur in the world. There are very few other considerations that could be relevant to interpreting them. Second, the way people see line drawings often suggests hypotheses about the represen- tations that human vision uses to express object geometry. Consider, for example, the phenomenon of pseudostability. Pseudostability is what typically happens when people first see drawings of ‘Impossible Objects”. Their initial impression is that they are seeing an object like any other. It takes time or prompting to register that the ‘object’ has properties which could not possibly co-exist. That immediately suggests that human impressions of object geometry are not expressed in terms which automatically ensure consistency (as a model or a complete quantitative description would). Ideas about the way they are expressed will be considered later. On the negative side, there are well known argu- ments against working intensively with line drawings. However, they are less compelling than they seem. One widespread view is that the processes involved in interpreting line drawings are a side issue; they cease to matter in the presence of information from binocular stereopsis, motion, shading, and other cues with a clearer physical basis. That is an eminently logical prediction, but it appears not to fit human vision. A wide range of studies indicate that methods of inferring 0262-8856/93/060342-11 0 1993 Butterworth-Heinemann Ltd image and vision computing 342