Correlative light-electron microscopy (CLEM) enables to relate dynamics (or functions) with structure for a better understanding of cell mechanisms. However, the LM and EM images are of very different size, spatial resolution, field of view, and appearance. Registration of LM and EM modalities is then a timely, important but difficult open problem, which still requires some manual assistance. We have designed an original automated CLEM retracing-and-registration method involving a common representation with an adaptive associated scale (or blurring), the determination of the EM patch geometry, and the specification of appropriate descriptors and similarity criterion for the EM patch search. Its efficiency is demonstrated on real CLEM images.