Image registration using power spectrum and cepstrum techniques

The use of power cepstrum analysis in image registration is explored. Rotational shifts and translational shifts are corrected separately. The technique involves two main ideas. First, after pre-processing to remove redundant information and information which could result in false registration parameters, a rotational shift is changed into a translational shift. Second, power cepstrum analysis is used to correct the translational shift. Because of the introduction of these ideas, this new algorithm can work very fast and accurately compared to conventional correlation techniques.

The primary objective of this algorithm is to compare fundus photographs taken at different times, thus making possible early detection of some retinal diseases. However, this algorithm also can be used for other applications, such as analyzing aerial photographs. Numerous pictorial examples illustrating the technique as applied to retinal photographs are included throughout this thesis.