Illumination Modeling and Normalization for Face Recognition
AMFG '03 Proceedings of the IEEE International Workshop on Analysis and Modeling of Faces and Gestures
A fast method of lighting estimate using multi-linear algebra
SINOBIOMETRICS'04 Proceedings of the 5th Chinese conference on Advances in Biometric Person Authentication
| Hi-index | 0.00 |
To deal with image variations due to illuminationproblem, Ramamoorthi and Basri have independentlyderived a spherical harmonic analysis for the Lambertianreflectance and linear subspace. Their theoretical workprovided a new approach for face representation,however both of them assume that the 3D surfacenormals and albedo (or unit albedo)are known, whichlimit this algorithm's application. In this paper, wepresent a novel method for modeling 3D face shape andalbedo from only three images and this work well fill theblank, which Ramamoorthi and Basri left. Our work isclosely related to photometric stereo, but conditions ofphotometric stereo for estimating albedo and surfacenormal are too strict to be applied for real application.Moreover, the conventionally used singular valuedecomposition (SVD) approach will lead to the notoriouslinear ambiguity and the solvent of this problem will needto introduce more constrains or more images. By takingthe advantage of similar 3D shape of all human faces, thehighlight of the new method is that it circumambulates thelinear ambiguity by 3D alignment. The experiment resultsshow that our estimated model can be perfectly employedto face recognition and 3D reconstruction.