Advanced free-form deformation and Kullback–Lieblier divergence measure for digital elevation model registration

Abstract

Registration is the process of transforming various images of the same object, place, etc. to confer one coordinate system, be they multimodal, multi-temporal of the same place or images of different places having certain common characteristics. Many methods have been tried for image registration; however, only a handful of methods may be said to work for digital elevation model (DEM) registration. The motivation of the present work is to perform a robust and efficient algorithm to perform Elevation model image registration. In this paper, we present a new approach for DEM registration, particularly for multimodal or multi-temporal DEMs. For time efficient, robust and precise registration of DEMs, a novel idea has been proposed using rational DMS-spline volumes (Xu et al. in J Comput Sci Technol 23(5):862–873, 1996; McDonnell and Qin in J Gr Tools 12(3):24–41, 2007) (the term DMS is acronym for the three authors, namely Dahmen et al. in Math Comput 59(199):97–115, 1997). This is based on the usage of arbitrary topology lattices (MacCracken and Joy in 23rd annual conference on computer graphics and interactive techniques proceedings. ACM-SIGGRAPH, pp 181–188, 1996; Feng et al. in Vis Comput Int Comput Gr 22(1):28–42, 2005). Using free-form deformation has lead to the usage of global and local transformations for registration of candidate image domain to reference image domain. Also, a similarity measure based on Kullback–Leibler divergence (KLD) has been used for measuring robustness of the method so proposed. The task of registration is achieved by minimizing the cost function using rational DMS-spline functions for local registration. After experimentations, the results show that the registration process, of registration of candidate DEM to reference DEM, could be completed successfully. Comparison of similarity measurement methods such as mutual information, correlation coefficient and peak signal-to-noise ratio with that of KLD-based has been performed. Comparative study with existing works suggests that the presented scheme is better, when compared with respect to above mentioned parameters.