Several studies in literature suggest hints of better shape estimations in complex scenes. We are testing the effects of scene complexity and ordinal transformations of luminance on perception of shape.

An observer judged the surface normal at 5 pre-marked points on computer rendered elliptic shaped objects under four different conditions: In a simple scene where the object is presented alone, in a complex scene where complexity is defined by number of ellipsoids present in the scene, and with two ordinal transformations which are power functions of intensity one compressive, one expansive. Stimuli were rendered stereo image pairs presented in a computer controlled Wheatstone stereoscope. All ellipsoids have the same surface color and surface reflectence function, which is Lambertian+specular (with specularity 0.001). Observer completed 400 trials for 2 different objects: 200 in each ordinal transformation: 50 trials for each of 4 scenes (2 simple and 2 complex). On each trial, observer adjusted a monocular gradient probe to estimate the normal of the surface at one of the 5 pre-marked points. For each of 5 pre-marked points, observer made 10 repetitions of settings. One observer completed the task.

In the range of ordinal transformations employed, we observe no discernible effect of these transformations on observer's judgment. Similarly, merely adding more objects having the same physical properties, such as color and BRDF, to the scene does not affect the observer's judgments. However, under certain conditions, such as an additional object in the scene which is casting shadow on the target object, observer's judgment is influenced substantially, especially around such a shadow boundary. Our findings suggest that human shape algorithm uses only ordering of luminance values in binocular shape from shading/contour/specularity and that a single ellipsoid is complex enough to permit stable reconstruction.