The assumed light direction for perceiving shape from shading

Abstract

Figure 1: The shading of an object's surface depends on the light source direction. The light is directed from the viewpoint in the left image, from 22 deg above the viewpoint in the middle image, and from 44 deg above the viewpoint in the right image. The object is positioned identically in each of the three views. In this paper, we present an experiment designed to test how shape perception is affected by changing the angle of the light direction. We found the lighting used in the center image led to the most accurate estimations of 3D shape. AbstractRecovering 3D shape from shading is an ill-posed problem that the visual system can solve only by making use of additional information such as the position of the light source. Previous research has shown that people tend to assume light is above and slightly to the left of the object [Sun and Perona 1998]. We present a study to investigate whether the visual system also assumes the angle between the light direction and the viewing direction. We conducted a shape perception experiment in which subjects estimated surface orientation on smooth, virtual 3D shapes displayed monocularly using local Lambertian shading without cast shadows. We varied the angle between the viewing direction and the light direction within a range +/-66 deg (above/below), and subjects indicated local surface orientation by rotating a gauge figure to appear normal to the surface [Koenderink et al. 1992]. Observer settings were more accurate and precise when the light was positioned above rather than below the viewpoint. Additionally, errors were minimized when the angle between the light direction and the viewing direction was 20-30 deg. Measurements of surface slant and tilt error support this result. These findings confirm the light-from-above prior and provide evidence that the angle between the viewing direction and the light direction is assumed to be 20-30 deg above the viewpoint.

A Perceptual-Statistics Shading Model

“…Follow-up studies indicated that this direction is from above the viewer and 12 • left from the vertical axis [33,20]. O'Shea and colleagues studied the assumed slant of the light direction on purely diffuse surfaces with no shadows [26]. They demonstrated that the surface slants were most accurate when the light source was 20 • − 30 • above the viewer.…”

“…In Figure 4, we illustrate an example of a bad and a good orientation of a gauge figure. This task has been employed for example by O'Shea et al to measure the accuracy of surface perception under varying slant of the illumination direction [26].Šoltészová et al utilized this test to compare the surface perception for different styles of shadow rendering [32]. Cole et al conducted a large-scale gauge-figure experiment, where they compared the accuracy of surface perception from automatic and man-made line-drawing representations of objects compared to their fully-shaded renderings [6].…”

“…The perception literature reports that the surface slant, as deduced from monoscopic renderings of 3D objects viewed on a screen, is systematically distorted, however there is no model representing this phenomenon [26,34]. The slant angle is understood as the angle between the surface normal and the viewing direction.…”

“…A model derived from statistical analysis of user evaluation has not been available before. It has been only attempted to model this effect as a parabolic function [26] or to use a simple shearing transformation in depth [34]. These approximations are consistent with the general expectation of perception but not founded on a statistical analysis of results of a perceptual study.…”

“…To demonstrate the effect of our approach, we applied our model to Lambertian shading and used purely diffuse-reflective materials. In all settings, the light source conforms to the assumed light direction [26]. Figure 1 shows a stream surface before (left-most) and after our modification (right-most).…”