Jbos Jean-Bernard Martens

An object-oriented model for brightness perception

An object-oriented model for the brightness perception of static images is presented. The philosophy behind the model is that the visual system aims at a brightness representation that displays object properties, and consequently is insensitive to variations in light source and changes in viewing position. The model assumes an ensemble of neural units that differ in receptive field size, i.e. identical operations are performed at a number of (spatial) scales. The brightness at each retinal position is the weighted sum of the neural activities that exist at this position in the different scales. The weighting function is such that the brightness impression is robust against variation in viewing distance. The operating characteristics of the neural units are derived from ganglion cell behaviour, and are robust against variation in light level. The model is able to unify different aspects of brightness perception, such as brightness induction and brightness assimilation. In order to gain in transparency, the approach adopted in this paper is a general one: we aim at showing how different aspects of brightness perception can be explained and unified within the model, at the expense of a less detailed description of psychophysical data.

Perceptual linearisation: Bridging the gap between simple and complex achromatic displays

Models of illusory and occluded contour and surface completion have focused on static spatial relations in planar surfaces. We consider several findings indicating that depth and motion play important roles in object completion. In one set of studies, effects of depth separation were studied in a speeded classification paradigm in which surface pieces were oriented in parallel or convergent (intersecting) planes. Subjects task was to respond “parallel” or “convergent” on each trial. It was hypothesised that object completion would facilitate performance, and that depth separation of the two visible areas would disrupt completion. Speed and accuracy were reliably reduced when object parts were separated in depth via stereoscopic information. In a second series of studies a speeded classification paradigm was used to test perception of illusory figures from edges specified solely by motion (accretion-deletion of texture). The aim of a third study was to test occluded object completion where the visible parts of the object were not simultaneously present, but revealed over time. Results showed reliable object completion effects when certain spatial and temporal relationships held among object parts. We discuss three elements of a revised theoretical model suggested by these results: (1) edge inputs to boundary completion processes need not be luminance edges—they may also be motion-specified edges; (2) the criterion of relatability governing edge interpolation processes must include depth relations as well as planar ones; (3) the criterion of relatability must be reformulated in terms of spatiotemporal, rather than merely spatial, relationships. Particulars of a new model that meets these conditions are briefly described.