K2 Summit and K2 Base Direct Detection Camera User's Guide
48
Addendum
Ensuring Validity of Defect Corrections
Defects in the camera, either in bias level and gain in Linear modes, or detection probability
and background counting rate in Counted modes, give rise to a fixed pattern in detected
images. These are patterns that recur in successive images and, in addition to introducing
noise, create difficulty with cross-correlation algorithms used for particle alignment and or
drift correction. Good reference images should effectively eliminate fixed pattern noise. One
way to test for the quality and effectiveness of reference images is to take two images, each
with half the dose normally used for the intended application and look at how the sum of the
two (which would include fixed pattern noise) compares with the difference (which would
not). The following script, which can be copied into the DM scripting window and run on the
two half-exposures yields a number, in percent, which predict how much the DQE will be
reduced (“derated”) by fixed pattern noise. A fixed pattern derating number of 4% would
mean that a DQE of 50% at half Nyquist would be reduced to 48% for an exposure of the
preferred length. It is important to use half exposures related to the intended exposure length
because the derating metric scales with exposure time due to the fact that while random noise
sources accumulate as the square root of exposure time, fixed pattern noise accumulates
linearly. This script can be used to confirm that the gain reference exposure factor has been
set high enough (see the section on K2 Summit Counted and Super-resolution gain reference
acquisition, page 24) and also to indicate when a fresh hardware dark reference needs to be
acquired (page 28).
//
//
Fixed pattern ratio - or exposure-weighted DQE derating by fixed pattern noise
//
//
Paul Mooney 10-16-12
//
// Instructions:
//
1.
Take two images of uniform illumination at 1/2 your total intended
exposure time at your planned dose-rate.
//
2.
Edit the margin variable to exclude an appropriate region around the edge
of the image.
//
3.
Run this script with the two images topmost in DM.
//
4.
The values in the image show the percent decrease in effective DQE at that
spatial frequency due to fixed pattern noise. For a 10s exposure (two five second
exposures) the FPN ratio should be under 10% but best under 4%.
//
5.
The ratio will scale linearly with exposure time.
//
6.
Reasons for higher values: gain reference taken with less than 100x
intended exposure time or aging dark reference.
//
7.
A pattern indicates dark reference issues. A flat elevation above zero
indicates gain reference issues.
//
image imga, imgb
gettwoimages("",imga, imgb)
string name = getname(imga)
number width, height, margin = .005
// fraction of image dimension
