A New 2.5D and 3D human face reconstruction approach for recogntion

摘要: Cet article presente une nouvelle approche de numerisation tridimensionnelle visages a base d\'un capteur stereoscopique actif. Elle est destinee la reconnaissance faciale, enrichie par des informations forme 3D. Apres etape d\'etalonnage en vue d\'estimer les parametres intrinseques et extrinseques du stereo, partie active (video projecteur) projette lumiere texturante permettant discrimination localisation sous-pixelique deux ensembles primitives sur images stereo. La mise correspondance ces se fait l\'aide notre adaptation programmation dynamique, ou, l\'exploitation geometrie epipolaire, associee d\'autres contraintes, permet reduire complexite probleme initialement bidimensionnel un monodimensionnel. L\'information profondeur est, consequent, retrouvee triangulation optique, suivie d\'etapes d\'interpolation maillage 3D d\'aboutir modeles partiels 2.5D visage. Finalement, operations registration, fusion mappage texture achevent reconstruction modele complet. Une evaluation nos resultats rapport ceux obtenus technologie laser est egalement presentee. In this paper, we propose complete and human face acquisition framework based on stereo sensor coupled with structured lighting source. We aim to develop an accurate low-cost solution dedicated the model-based recognition techniques (FRT). First, calibrate in order estimate its optical characteristics geometrical parameters (the offline phase). Second, epipolar geometry projection of special light online capture phase), improves resolution matching problem, by transforming it into one-dimensional search problem sub-pixel features matching. Next, apply our adapted optimized dynamic programming algorithm pairs which are already located each scanline. Finally, information is found computing intersection rays coming from pair matched features. The achieved face model produced pipeline four steps: (a) Spline-based interpolation, (b) Partial models\' alignment then integration, (c) Mesh generation, (d) Texture mapping. present, furthermore, some results discuss both qualitative quantitative evaluations proposed scheme. Keywords : Vison active, visages, appariement Splines cubiques, faciale.; Active modeling, matching, cubic Splines, recognition. Journal Sciences Pour l\'Ingenieur. Vol. 7 2006: pp. 37-44