Particle picking algorithm provides us a set of particle temp/zeyun, each of which
corresponds to one projection of the 3D particle structure. However, some of the particles
may have roughly the same "appearance", which means they differ from each other only
under certain amount of in-plane rotation, translation and/or scaling. All particles with
similar "appearance" can thus be aligned and averaged to obtain an averaged image that is
supposed to have a much better signal-to-noise ratio. But before we do the alignment and
averaging, we must classify all the particle temp/zeyun into different groups based upon their
"appearance". All such averaged temp/zeyun (one per class) are used as inputs to the Reconstruction. The following picture shows the pipeline of
particle classification, alignment and averaging, starting from the detected particles and
ending up with a set of particle averages.
|Fig. 1 Pipeline of particle classification.|
Image classification is a well-studied topic in image processing. However, very few techniques
on particle classification have been explored in single particle reconstruction. We intend to
do something on this problem but we have not yet started.