Image Registration by Minimizing Tsallis Divergence Measure

In this paper, a novel image registration method is proposed which makes use of the a priori knowledge learned from pre-aligned training images. Two images are registered if the difference between the observed joint distribution estimated from them and the expected joint distribution obtained from the aligned training images is minimized. The difference is measured by the Tsallis divergence measure. The performance of the new method is compared with the classical Shannon mutual information and Tsallis mutual information. Experimental results show that the proposed method is computationally more efficient with higher registration accuracy and faster registration convergence.