Correlation-based 2D registration method for single particle cryo-EM images

Abstract

The amount of image data generated in single particle cryo-electron microscopy (cryo-EM) is huge. This technique is based on the reconstruction of the 3D model of a particle using its 2D projections. The most common way to reduce the noise in particle projection images is averaging. The essential step before the averaging is the alignment of projections. In this work, we propose a fast 2D rigid registration approach for alignment of particle projections in single particle cryo-EM. We used cross-correlation in Fourier domain combined with polar transform to find the rotation angle invariant to the shift between the images. For translation vector estimation we used a fast version of upsampled image correlation. Our approach was evaluated on specifically created synthetic dataset. An experimental comparison with a widely used in existing software iterative method has been performed. In addition, it was successfully applied to a real dataset from the Electron Microscopy Data Bank (EMDB).