Intensity-based point-spread-function-aware registration for multi-view applications in optical microscopy

We present an algorithm to spatially register two volumetric datasets related via a rigid-body transform and degraded by an anisotropic point-spread-function (PSF). Registration is necessary, for example, when fusing data in multi-view microscopy. Automatic algorithms that only rely on maximizing pixel similarity, without accounting for the anisotropic image formation process, provide poor results in such applications. We propose to solve this problem by re-blurring the reference and test data with transformed forms of the PSF, in order to make them comparable, before minimizing the mean squared intensity difference between them. Our approach extends the pyramid-based sub-pixel registration algorithm proposed by Thévenaz et al., 1998, that employs an improved form of the Marquardt-Levenberg algorithm. We show, via simulations, that our method is more accurate than the conventional approach that does not account for the PSF. We demonstrate our algorithm in practice by registering multi-view volumes of a zebrafish larva acquired using a wide-field microscope.