Abstract
This manuscript presents a new approach to dense stereo matching pro- blem. One of the main originality of this work is related to the use of the fuzzy set theory to replace precise similarity measure by imprecise resemblance measure in the matching process. This new matching measure is more robust to sampling and quan- tization due to a new modelling of a local digital luminance measure called ”fuzzy pixel”. This modelling associates to each pixel of the image a fuzzy domain and a fuzzy classification of the luminance. Our method associate to each pixel of the first image a fuzzy domain of the second image based on luminance and epipolar information. Each domain is modified by an iterative and cooperative process in order to provide new domains that respect geome- tric constraint like ordering and uniqueness. The validity of our approach is illustrated with experimentations on synthetic and real stereoscopic images.