CLASSES

 

Reference-free classification of images.


Usage
classes <footprints> <# classes> [list] [weight=<file>]
 
<footprints> file containing 'footprints' of original images. 'footprint' file contains filename of the original image file it was generated from.
<#classes> number of classes to generate. A good starting value is the number of images divided by something between 10 and 50.
list causes a list file to be generated describing which images went into each class.
weight=<file> not usually used. A weight file can be used to specify the relative importance of each footprint variable.
output files Generates a set of files called cls###.hed

Typical Usage
classes start.fp 50

Description
This program is used to classify images when no preliminary 3d model is available. It uses a set of 'footprints' generated by footprint[2,3] to judge the relative similarity of individual images and split them into as many classes as you like. Classification is done using the k-mean method.

Once the images have been classified, typically classalignall will be used to generate class averages. Only a small fraction of the class averages will typically represent good particle views, but it should generate enough that you can generate a preliminary model to use for 3d refinement.
 

EMAN Copyright 1997-8 Steve Ludtke, Phil Baldwin