ccdmask

ccdproc.ccdmask(ratio, findbadcolumns=False, byblocks=False, ncmed=7, nlmed=7, ncsig=15, nlsig=15, lsigma=9, hsigma=9, ngood=5)

Uses method based on the IRAF ccdmask task to generate a mask based on the given input.

Note

This function uses lines as synonym for the first axis and columns the second axis. Only two-dimensional ratio is currently supported.

Parameters:

ratioCCDData

Data to used to form mask. Typically this is the ratio of two flat field images.

findbadcolumnsbool, optional

If set to True, the code will search for bad column sections. Note that this treats columns as special and breaks symmetry between lines and columns and so is likely only appropriate for detectors which have readout directions. Default is False.

byblocksbool, optional

If set to true, the code will divide the image up in to blocks of size nlsig by ncsig and determine the standard deviation estimate in each block (as described in the original IRAF task, see Notes below). If set to False, then the code will use scipy.ndimage.percentile_filter to generate a running box version of the standard deviation estimate and use that value for the standard deviation at each pixel. Default is False.

ncmed, nlmedint, optional

The column and line size of the moving median rectangle used to estimate the uncontaminated local signal. The column median size should be at least 3 pixels to span single bad columns. Default is 7.

ncsig, nlsigint, optional

The column and line size of regions used to estimate the uncontaminated local sigma using a percentile. The size of the box should contain of order 100 pixels or more. Default is 15.

lsigma, hsigmafloat, optional

Positive sigma factors to use for selecting pixels below and above the median level based on the local percentile sigma. Default is 9.

ngoodint, optional

Gaps of undetected pixels along the column direction of length less than this amount are also flagged as bad pixels, if they are between pixels masked in that column. Default is 5.

Returns:

masknumpy.ndarray

A boolean ndarray where the bad pixels have a value of 1 (True) and valid pixels 0 (False), following the numpy.ma conventions.

Notes

Similar implementation to IRAF’s ccdmask task. The Following documentation is copied directly from: http://stsdas.stsci.edu/cgi-bin/gethelp.cgi?ccdmask

The input image is first subtracted by a moving box median. The median is unaffected by bad pixels provided the median size is larger that twice the size of a bad region. Thus, if 3 pixel wide bad columns are present then the column median box size should be at least 7 pixels. The median box can be a single pixel wide along one dimension if needed. This may be appropriate for spectroscopic long slit data.

The median subtracted image is then divided into blocks of size nclsig by nlsig. In each block the pixel values are sorted and the pixels nearest the 30.9 and 69.1 percentile points are found; this would be the one sigma points in a Gaussian noise distribution. The difference between the two count levels divided by two is then the local sigma estimate. This algorithm is used to avoid contamination by the bad pixel values. The block size must be at least 10 pixels in each dimension to provide sufficient pixels for a good estimate of the percentile sigma. The sigma uncertainty estimate of each pixel in the image is then the sigma from the nearest block.

The deviant pixels are found by comparing the median subtracted residual to a specified sigma threshold factor times the local sigma above and below zero (the lsigma and hsigma parameters). This is done for individual pixels and then for column sums of pixels (excluding previously flagged bad pixels) from two to the number of lines in the image. The sigma of the sums is scaled by the square root of the number of pixels summed so that statistically low or high column regions may be detected even though individual pixels may not be statistically deviant. For the purpose of this task one would normally select large sigma threshold factors such as six or greater to detect only true bad pixels and not the extremes of the noise distribution.

As a final step each column is examined to see if there are small segments of unflagged pixels between bad pixels. If the length of a segment is less than that given by the ngood parameter all the pixels in the segment are also marked as bad.