Automatic image registration by stochastic optimization of mutual information

Image registration is the process by which we determine a transformation that provides the most accurate match between two images. The search for the matching transformation can be automated with a suitable choice of metric, but can be very time-consuming and tedious. We consider a registration algorithm that combines a simple yet powerful search strategy based on optimization of mutual information using a stochastic gradient, together with a wavelet-based multi-resolution pyramid. It is tested using a pair of fundus eye images, which is matched using a six-parameter affine transformation. This extends previous work based on the three-parameter transformation (Cole-Rhodes, A. et al., SPIE Aerosense 2002, Wavelet Applications IX, 2002). The registration algorithm is implemented in a multiresolution manner using a wavelet pyramid.