A. Quddus, O. Basir, and F. Karray (Canada)
Wavelets, Image Registration, Mutual Information, Fuzzy Logic, Medical Imaging.
In this report, a novel, multiscale wavelet-based image reg istration technique is proposed for image retrieval appli cations. The proposed technique addresses the problem of registration and alignment in the integrated framework. The use of multiscale wavelet representation with mutual information (MI) is expected to facilitate matching of im portant anatomical structures at multiple resolutions. We propose to use a dyadic grid in the parameter search space for efficient computation in the multiscale domain. We pro pose to use fuzzy logic based control to adaptively change the search step size based on rate of change in MI at coarse scale. The use of fuzzy logic is justified due to two impor tant reasons. First, mathematical modeling of change in MI in various image modalities is difficult to obtain. Second, implementation of fuzzy control doesn’t have any computa tional overhead as it is realized using simple look-up tables. The proposed approach has several novel aspects including the use of MI in multiscale wavelet domain and the use of fuzzy control for efficient parameter search in multiscale domain. The technique is also tested under varying noise levels. The preliminary results show the efficacy of the ap proach.
Important Links:
Go Back
