Item

Abstract

Many materials — such as wax, glass, fruit flesh, and human skin — transmit as well as reflect light. When light passes through an object, it gives the object a characteristic appearance of transparency or translucency. How do we identify materials that transmit light? What are the perceptual dimensions of translucency? What image properties make an object look transmissive rather than opaque? We use a combination of ecological optics and psychophysics to address these questions. Almost all research on the perception of transmissive materials is based on Metelli's linear ‘episcotister’ model. Unfortunately, the episcotister is a poor physical model of light transport in real materials. It ignores: (i) specular reflection, (ii) surface ‘frosting’ (as in ‘frosted’ glass), (iii) refraction and (iv) sub-surface scatter. These shortcomings have profound consequences for the perception of translucency. They affect both the perceptual parameters of transparent materials and the image properties that make objects look transmissive. Recent advances in computer graphics (Jensen et al. SIGGRAPH, 2001) allow us to simulate translucent materials realistically. We can now systematically vary sub-surface light scatter, while holding shape, lighting and viewpoint constant. This allows us to identify image properties that are diagnostic of translucency. We discuss how image statistics based on luminance, contrast, orientation, and scale contribute to the perception of translucency, as well as other cues such as highlights and shadows. Using psychophysical matching tasks, we measure how refractive index and translucency appear to change as the object or lights are moved. Importantly, we find that traditional cues to transparency (e.g. X-junctions and visibility of the underlying layer) are not necessary, and in fact rarely occur in realistic images of transparent objects. Indeed, subjects can estimate an object's translucency even when it is floating in a featureless void.