Demon registration for 3D images obtained by serial block face scanning electron microscopy

In the last decade “Demon”-based matching algorithms have largely demonstrated their efficiency in non-rigid registration of clinical and biomedical images. This paper present an experimental study of a symmetrized variant of Thirion's demons algorithm applied to the realignment of 3D images obtained by serial block face scanning electron microscopy (SBFSEM). Because demons are sensible to local intensity difference, it is necessary to ensure that both images have matching intensity distributions. We propose to perform the demon registration on an image intensity transformation associating total-variation denoising and modified contrast-limited adaptive histogram equalization. We compare the performance of the presented method with a B-spline based free form deformation method on a SBFSEM stack. This preliminary study shows that the demon registration applied to the proposed image intensity transformation is a fast and robust algorithm. Besides it always exhibit a better compromise in term registration accuracy and smoothness of the transformation field than the B-spline method.