作者: Q.-Tuan Luong , Pascal Fua , Yvan Leclerc
关键词: Artificial intelligence 、 Photometric stereo 、 Multiple view 、 Complement (set theory) 、 Image (mathematics) 、 Computer vision 、 Surface (mathematics) 、 Computer science 、 Stability (probability) 、 Real image
摘要: We introduce a new methodology for radiometric reconstruction from multiple images. It opens possibilities because it allows simultaneous recovery of varying unknown illuminants (one per image), surface albedoes, and cameras' responses. Designed to complement geometric techniques, only requires as input the geometry scene cameras. Unlike photometric stereo approaches, is not restricted images taken single viewpoint. Linear non-linear implementations in Lambertian case are proposed; simulation results discussed compared related work demonstrate gain stability; on real shown.