Shape from shading from images rendered with various surface types and light fields

Shape constancy is referred to as the tendency for the perceived shape of an object to remain unchanged even under changed viewing and illumination conditions. We investigated, in two experiments, whether shape constancy would hold for images of 3-D solid objects defined by shading only, whose renderings differed in terms of surface material type (bi-directional reflectance distribution functions), light field, light direction, shape, and specularity. Observers were presented with the image of a sphere or an ellipsoid and required to set perceived orientation and cross-section profile on designated points of the image. Results showed that shape judgments varied with all the aforementioned variables except specularity. Shape estimates were more precise with specular than asperity scattering surfaces, collimated than hemispherical diffuse lighting conditions, lower than higher elevations, spherical than ellipsoidal shapes, but not different between surfaces having differing specularity. These results suggest that shape judgments are made largely on the basis of the overall intensity distribution of shading, and that the portions of intensity distribution that are due to nonstructural variables such as surface material type or light field are not excluded in the process of shape estimation, as if being due to structural components. It is concluded that little constancy is expected in the perception of shape from shading.     

Visually controlled matching of pattern movement

Utrecht Biophysics Research Institute, Rijksuniversiteit Utrecht, Utrecht, The Netherlands Subjects were asked to match the speeds of two moving random-dot patterns seen through circular apertures. The speed of one pattern that moved horizontally toward the right of a computer screen changed continuously. The speed of this pattern represented the target. It was to be matched with the speed of the second pattern, which moved in the opposite direction. The subject controlled the speed of the second pattern by means of an isometric joystick. The distance between the apertures on the screen as well as the subject’s distance from the screen served as experimental parameters. In this way, the effects of both spatialand temporal transients of pattern speed on human tracking performance were studied. To avoid anticipation by the subject, the amplitude and the frequency of the target pattern speed changed pseudorandomly. The accuracy with which the subject performed the matching task was influenced by the mean pattern speed and the parameters of the visual field. Within lower velocity ranges, the subject’s sensitivity to the instantaneous speed differences varied according to Weber’s law. The cross-correlation of the velocity time courses decreased when the mean speed of the target pattern was increased. Two stimulus parameters had a strong influence on the modulation of the correlation value: (1) the angular size of the stimulus on the retina and (2) the retinal eccentricity of the stimulus.     

Haptic perception of parallelity in the midsagittal plane.

Previous studies [Perception 28 (1999) 1001; Perception 28 (1999) 781] on the haptic perception of parallelity on a horizontal plane showed that what subjects haptically perceive as being parallel deviates considerably from what is physically parallel. The deviations could be described with a subject-dependent orientation gradient in the left-right direction. The gradients found in the bimanual conditions were significantly larger (about 70%) than those in the unimanual conditions. The questions to be answered in the present study are the following: (1) Does the haptic perception of parallelity in the midsagittal plane also show systematic deviations from veridicality? (2) Are the unimanual and bimanual performances again quantitatively but not qualitatively different? The set-up consisted of a plate positioned in the midsagittal plane of the subject. The subject touched the right side of the plate with his/her right hand and the left side with the left hand. The results show again large systematic deviations. The major part of the deviations can be described by means of a subject-dependent orientation gradient in the vertical direction. The quantitative (but not qualitative) difference between the unimanual and the bimanual conditions is much larger in the midsagittal plane than in the horizontal plane.     

An antisymmetric psychometric function on a logarithmic scale

This very brief report introduces a psychometric function, very suitable for psychophysical data that displays Weber-like behaviour, because it is antisymmetric on a logarithmic scale.     

Effects of context on a visual 3-D pointing task

We examined the effects of egocentric and contextual references on a 3-D exocentric pointing task. Large systematic deviations were found for the slant (angle in the horizontal plane). For most observers, the deviations were smaller when the veridical pointing direction was parallel to a wall. For some observers the size of the deviations was also dependent on whether the veridical pointing direction was frontoparallel or not. For the tilt (angle in the vertical plane), the deviations were smaller and less systematic. Hence, although observers show comparable systematic deviations, the way in which the presence of structure in an environment is used for judging positions of objects is observer-dependent.     

The visual light field

Human observers are sensitive to the '(physical) light field' in the sense that they have expectations of how a given object would appear if it were introduced in the scene in front of them at some arbitrary location. Thus the 'visual light field' is defined even in the 'empty space' between objects. In that sense the light field is akin to visual space considered as a 'container'. The visual light field at any given point can be measured in psychophysical experiments through the introduction of a suitable 'gauge object' at that position and letting the observer adjust the appearance of that gauge object (eg through suitable computer rendering) so as to produce a 'visual fit' into the scene. The parameters of the rendering will then be considered as the measurement result. We introduced white spheres as gauge objects at various locations in stereoscopically presented photographic scenes. We measured the direction ('direction of the light'), diffuseness ('quality of the light' as used by photographers and interior decorators), and intensity of the light field. We used three very different scenes, with very different physical light fields. The images were geometrically and photometrically calibrated, so we were in a position to correlate the observations with the physical 'ground truth'. We report that human observers are quite sensitive to various parameters of the physical light field and generally arrive at close to veridical settings, although a number of comparatively minor systematic deviations from veridicality can be noted. We conclude that the visual light field is an entity whose existence is at least as well defined as that of visual space, despite the fact that the visual light field hardly appears as prominently in vision science as it does in the visual arts.     

Detection of the sign of expansion as a function of field size and eccentricity

Small optical flow fields are involved in object and shape recognition. These tasks depend on high resolution and low speed of the flow-field elements. Large optical flow fields are involved in tasks such as orientation and navigation, which require flow-field elements to move at high speed. From the above, we would expect to find different behavior of subjects for different parameter ranges of field size and speed of the elements. In this paper, we address the question of whether such different behavior exists for a task that is associated neither with object recognition nor with navigation. We obtained detection thresholds for expansion in the presence of translation for a wide range of field sizes. The same paradigm was used to investigate whether subjects made efficient use of peripheral information. We found viewing-distance invariance, meaning that subjects’ performance scaled with stimulus size. Subjects performed very similarly with and without foveal information.