After a 15-year preparatory work, a group of mathematicians and computer scientists is ready to reconsider and formalize the main steps of image processing, and to devise a way to make them fully automatic. The project will conceive, revise or accelerate the dozen highly accurate algorithms necessary to establish a universal image processing chain, applicable to all raw digital images obtained from reflex or compact off-the-shelf cameras, and to all more specialized image generation systems, including a digital photon capture array (CCD or CMOS,...). The direct applications will be: - a fully autonomous image processing chain transforming any raw image into a distortion corrected and noise free visible image with optimal color and contrast. On line demo and C-code will be made available to the community; - a complete 3D image reconstruction system for an in-project Earth observation satellite (MISS project, Centre National d'Etudes Spatiales). This Earth scanner will take stereo pairs of the Earth at 40 centimeters resolution, permitting to build highly accurate urban elevation models. - a camera based passive 3D reconstruction system with $\pi/40000$ angular precision competing with the best active short range 3D triangulation scanners. Getting a reproducible 1/20 pixel accurate image processing chain from raw photon sensor data to 3D reconstruction with fully unsupervised algorithms will be a major change of the discipline. The team has devised a new image statistical theory and invented several new mathematical processes dealing with image noise and projective invariant image matching. The team has strongly contributed to the world's most successful image processing company, DxO Labs, and to the image processing line of three Earth observation satellites.