Starting from a given time-migrated zero-offset data volume and time-migration velocity, recent literature has shown that it is possible to simultaneously trace image rays in depth and reconstruct the depth-velocity model along them. This, in turn, allows image-ray migration, namely to map time-migrated reflections into depth by tracing the image ray until half of the reflection time is consumed. As known since the 1980s, image-ray migration can be made more complete if, besides reflection time, also estimates of its first and second derivatives with respect to the time-migration datum coordinates are available. Such information provides, in addition to the location and dip of the reflectors in depth, also an estimation of their curvature. The expressions explicitly relate geological dip and curvature to first and second derivatives of reflection time with respect to time-migration datum coordinates. Such quantitative relationships can provide useful constraints for improved construction of reflectors at depth in the presence of uncertainty. Furthermore, the results of image-ray migration can be used to verify and improve time-migration algorithms and can therefore be considered complementary to those of normal-ray migration. So far, image-ray migration algorithms have been restricted to layered models with isotropic smooth velocities within the layers. Using the methodology of surface-to-surface paraxial matrices, we obtain a natural extension to smooth or layered anisotropic media.