IPP Software Navigation Tools IPP Links Communication Pan-STARRS Links

Changes between Version 1 and Version 2 of Stack_Algorithm


Ignore:
Timestamp:
Oct 25, 2013, 4:17:04 PM (13 years ago)
Author:
watersc1
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • Stack_Algorithm

    v1 v2  
    33== Input selection ==
    44
     5Each input warp has an associated PSF file.  These PSFs are read, and the associated FWHM value is stored for each image.  These values are then used to determine which warps to use in constructing the stack.  MD field stacks have a variety of different cuts depending on if the final stack is being optimized for depth or image quality.  For ThreePi stacks, the following methods are used.
     6
     7In PV1, the warp rejection was performed in two passes.  First, any warp with input FWHM greater than a specified hard limit maxFWHM (10 pixel) was rejected.  Following this, the ensemble mean and sigma were calculated, and any warp with FWHM greater than a clip threshold, clipFWHMnSig, (1.5 x sigma) were rejected as well.  This algorithm was rather harsh, and was poorly tuned for situations in which the distribution of warp FWHM values was not unimodal.
     8
     9PV2 has introduced a new rejection algorithm that was designed to handle such cases better.  The first step is to determine a minimum and maximum value.  The minimum value is the "threshFWHM", a value below which no input can be rejected.  The maximum value is the maxFWHM, and corresponds to the hard limit in the PV1 algorithm.  At this point, a clipped rejection can be used to mimic PV1 stacks, although the threshFWHM is still respected.  The alternative is to use a mixture-model analysis.  This attempts to determine if the distribution of FWHM values is best represented by a single Gaussian population or by a combination of two Gaussians.  If the distribution is unimodal (Puni > 0.5), or contains a small number of values (Ninput <= 4), the maxFWHM value is used as the upper limit.  If the distribution is bimodal, the limit is set as equal to mean_popular + clipFWHMnSig * sigma_popular, where the "popular" statistics are selected for the Gaussian with the largest fraction of the input values.  This limit is again checked against threshFWHM and maxFWHM, and used to reject inputs.
     10
     11The PV2 value of maxFWHM is set to 10 pixel, which retains many more inputs on average than the PV1 algorithm.  This may be revised for PV3, but has been kept high to ensure that the final stacks are constructed with the most inputs.
    512== PSF Matching ==
    613