The contribution of motion to shape-from-specularities

Roby Michelangelo Vota, Nahide Dicle Dövencioglu, Ohad Ben-Shahar, Katja Dorschner, Maarten Wijntjes

Research output: Contribution to journalArticlepeer-review


To infer the 3D shape of an object, the visual system often relies on 2D retinal input. For ideal
specular surfaces, the retinal image is a deformation of the surrounding environment. Since many
configurations of shape and environment can potentially generate the same retinal image, 3D
specular shape estimation is ambiguous. A relative motion between object, observer and
environment produces dynamic information on the retina, called specular flow. From a
computational point of view, this specular flow may diminish perceptual ambiguities. For this
research, two novel, smooth shapes were rendered with two different environment maps (a
forest and a city) and under two motion conditions (static and dynamic). In the dynamic
condition, the surface and the observer were kept relatively static, but the surrounding
environment map was rotated at sinusoidal speed around the vertical axis, generating ‘flowing’
reflections on the surface. Eight observers performed the gauge figure task (attitude probe) with
these stimuli. The analysis of variance indicated that both inter-observer correlations and
correlations with the 3D input model were higher for the static presentation than for the
dynamic one. Results imply that specular flow, despite offering a computational advantage, is
not beneficially used by the human visual system.
Original languageEnglish GB
Pages (from-to)352
StatePublished - 2015


Dive into the research topics of 'The contribution of motion to shape-from-specularities'. Together they form a unique fingerprint.

Cite this