A new hybrid registration technique for dental panoramic X-ray images

Abstract

This paper presents a fast and fully automatic hybrid image registration technique using wavelet-based hierarchical approach. At low resolution level, mutual-information (MI) based registration using similarity transformation model is performed. At high resolution level, scale invariant feature transform (SIFT) based registration is applied to accelerate registration convergence and to achieve good computational efficiency using rough parameters extracted from MI. To remove outliers automatically a RANdom Sample And Consensus (RANSAC) algorithm is applied. A comparison between proposed technique with hybrid registration approach using MI and SIFT in the spatial domain and with three wavelet-based registration techniques is achieved: point mapping registration, SIFT-based registration, and MI-based hierarchical registration. The quality of the registration process was measured using the following criteria: normalized cross-correlation coefficient (NCCC), percentage relative root mean square error (PRRMSE), and run time. The application of the selected techniques to dental panoramic X-ray images has shown that proposed wavelet-based approach combining MI, SIFT, and RANSAC algorithm gives the best results and can be used efficiently for registration of X-ray images. It gave 0.7805 NCCC, 0.1040% PRRMSE, and 22 seconds run time.