In this paper, we present a new method for ultrasound image registration. For each image to be registered, our method first applies an ultrasound-specific information-theoretic feature detector, which is based on statistical modeling of speckle and provides a feature image that robustly delineates important edges in the image. These feature images are then registered using differential equations, the solution of which provides a locally optimal transformation that brings the images into alignment. We describe our method and present experimental results demonstrating its effectiveness, particularly for low contrast, speckled images. Furthermore, we compare our method to standard gradient-based techniques, which we show are more susceptible to misregistration.
Zhe Wang, Gregory G. Slabaugh, Gozde B. Unal, Tong