Photo-geometric registration via a coupled L∞ filter

Presented is a coupled L_∞ filter capable of simultaneously estimating model parameters for photo-geometric registration between two (or more) images. We also introduce a colour code based clustering technique to group features for fast and reliable matching. This technique is based on the CIE colour model and is invariant to illumination changes. An increase in speed of almost 2 times is observed when constructing image mosaics from colour codes. We show that our filter estimates both the geometric and photometric models at the same time with low projection error. Furthermore, we show that our filter outperforms non-linear optimisation techniques.