bUnwarpJ: Consistent and Elastic Registration in ImageJ. Methods and Applications

Image registration is a common problem in computer vision, and speci cally in biomedical applications that require inter or intra modality image alignment. We have developed an ImageJ plugin called bUnwarpJ, which elastically registers pairs of images. This simple and easy-to-use plugin can be used by researchers and clinicians to create anatomical atlases, segment images using atlases, align pairs of images distorted by both physical and acquisition related distortions, etc. Registering two images consists on nding the image transformation that maps corresponding pairs of pixels between the original and "distorted" images. We use here the term distorted in a wide sense, to account not only for "sensu strictu" image distorsions, but also for anatomical variations between or within individuals. We use an algorithm that simultaneously calculates the direct and inverse transformations and minimizes the similarity error between the target and source images after imposing a consistency constraint. This approach provides bidirectional registration from image A to B or from B to A in a single computation. We use B-splines to represent both images and deformations and make use of a powerful optimizer to converge fast to the best image alignment. Our plugin allows guiding the registration process using the image similarity, the consistency of the deformations, vector-spline regularization and/or a set of optional landmarks, which can be calculated and fed from other ImageJ plugins such as the automatic extractors Scale-Invariant Feature Transform (SIFT) and Multi-Scale Oriented Patches (MOPs). The user can give a weight to each of these terms in the registration process. This paper provides a general description of the algorithm and its implementation in order to help developers and users to exploit all its potential.