Registration of digital ophthalmic images using vector mapping

This paper outlines a technique for the automated alignment of digital images of the retina under the assumption that they differ spatially by a geometric global transform. This alignment is a form of image registration. The solution is based on generating vectors from known, predominantly common points in the images to be registered. All potential transforms between the vectors are generated, with the correct registration producing a tight cluster of data points in the space of transform coefficients. This cluster is identified using the expectation maximization algorithm, and the optimum global transform calculated. The technique is applied to two types of retinal image-optical photographs and fluorescein angiograms, demonstrating that it can be readily used to provide cross-modal registration. An objective analysis of this system is also outlined, and examples of an application to real clinical images presented.