Image registration robust to sparse large errors

Registration is difficult when images to be registered contain sparse but large-valued differences. We present a method for robust registration that ignores some fraction of large differences, while constraining the sparseness of these errors. We apply the method to stabilize microscopy videos of C. elegans tissues, in which bright moving filaments and tissue wounding appear as sparse large-valued differences. We demonstrate the advantage of the method on both synthetic and real data compared to state-of-the-art methods.