Skip to main content

Advancing the theory of photometric stereo for texture extraction

Objective

The main goals of the project are to advance the theory of photometric stereo, to develop new practical algorithms for robust photometric stereo, and to broaden the limits of photometric stereo applicability and accuracy for the specific case of surface te xture recovery. Surface texture is modelled as a composition of surface geometry (a bump map) and first-order reflectance characteristics (a Lambertian albedo map). This is why photometric stereo is selected as the method of interest, since it excels in re covery of these two entities. Its input is images acquired under a fixed viewpoint and changing illumination. The motivation for detailed photometric stereo analysis comes from three main points: 1. Applicability. It is usually assumed that the illuminatio n conditions used to acquire individual input images are known. If this is not the case, the output entities can be recovered only up to a global ambiguity containing nine degrees of freedom. Recent work by the applicant shows that when an ideal specular c omponent of reflectance is present, the ambiguity reduces to only two degrees of freedom. We intend to extend this result for more complex reflectance models. 2. Robustness. For high-quality results, common image phenomena such as specular interreflections , cast shadows, etc. must be accounted for, otherwise the results are biased. Such phenomena represent outliers to the assumed model of reflectance. We need to analyse the possibilities for fast and robust methods for their detection. 3. Accuracy. While Po int 2. addresses the phenomena that cause gross deviations of image intensities from the values predicted by a reflectance model, this point focuses on more tiny phenomena of diffuse interreflections in the concavities of a bump map. When the illumination is oblique, shadowing effects further complicate the problem. We will address these issues by explicitly accounting for interreflections and shadows in the algorithm, which we believe will give us significantly better estimates of bump and albedo maps.

Call for proposal

FP6-2002-MOBILITY-5
See other projects for this call

Coordinator

HERIOT-WATT UNIVERSITY
Address
Riccarton
EH14 4AS Edinburgh
United Kingdom