Colour Meets Geometry in Colorimetric Filter Design

Panoramas are formed by stitching together two or more images of a scene viewed from different positions. Part of the solution to this stitching problem is ‘solving for the homography’: where the detail in one image is geometrically warped so it appears in the coordinate frame of another. In this paper, we view the spectral loci for a given camera and the human visual system (i.e. their respective chromaticity diagrams) as two pictures of the same ‘scene’ and warp one to the other by finding the best homography. When this geometric distortion renders the two loci to be identical then there exists a unique colour filter (that falls gracefully from the derivation without further calculation) which makes the camera colorimetric (the camera+filter measures RGBs that are exactly linearly related to XYZs). When the best homography is not exact the filter derived by this method still makes cameras approximately colorimetric. Experiments validate our method.