Environment Matting into Photographs using Coloured RTcams

There has been continuing interest in using environment matting to capture the real-world's reflection and refraction from a foreground object. Environment matting require a calibrated warp, which is often defined non-parametrically. Although this is a powerful approach it suffers from two basic limitations: (i) there is a "blind spot" that cannot be a calibrated; (ii) a slight change in the object requires a full recalibration.This paper introduces a parametric approach that can overcome both of these limitations. We build upon a generalised model of a non-linear camera called an RTcam (Rational Tensor Camera, introduced in Section 3), for the first time showing its value to environment matting. In addition, we show how a restriction on RTcams can simulate the change in colour of an object as it appears in a coloured reflector (refractor). Hence we account for both geometric and colour changes with a single mathematical model. Finally, we show how, using the same camera model, the reflector properties can mapped from one to another.