ccdmask: Create bad pixel mask from CCD flat field images

Package: ccdred

Usage

ccdmask image mask

Parameters

image: CCD image to use in defining bad pixels. Typically this is a flat field image or, even better, the ratio of two flat field images of different exposure levels.

mask: Pixel mask name to be created. A pixel list image, .pl extension, is created so no extension is necessary.

ncmed = 7, nlmed = 7: The column and line size of a 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.

ncsig = 15, nlsig = 15: 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.

lsigma = 6, hsigma = 6: Positive sigma factors to use for selecting pixels below and above the median level based on the local percentile sigma.

ngood = 5: Gaps of undetected pixels along the column direction of length less than this amount are also flagged as bad pixels.

linterp = 2: Mask code for pixels having a bounding good pixel separation which is smaller along lines; i.e. to use line interpolation along the narrower dimension.

cinterp = 3: Mask code for pixels having a bounding good pixel separation which is smaller along columns; i.e. to use columns interpolation along the narrower dimension.

eqinterp = 2: Mask code for pixels having a bounding good pixel separation which is equal along lines and columns.

Description

Ccdmask makes a pixel mask from pixels deviating by a specified statistical amount from the local median level. The input images may be of any type but this task was designed primarily for detecting column oriented CCD defects such as charge traps that cause bad columns and non-linear sensitivities. The ideal input is a ratio of two flat fields having different exposure levels so that all features which would normally flat field properly are removed and only pixels which are not corrected by flat fielding are found to make the pixel mask. A single flat field may also be used but pixels of low or high sensitivity may be included as well as true bad pixels.

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.

The bad pixel mask is created with good pixels identified by zero values and the bad pixels by non-zero values. The nearest good pixels along the columns and lines for each bad pixel are located and the separation along the columns and lines between those pixels is computed. The smaller separation is used to select the mask value. If the smaller separation is along lines the linterp value is set, if the smaller separation is along columns the cinterp value is set, and if the two are equal the eqinterp value is set. The purpose of this is to allow interpolating across bad pixels using the narrowest dimension. The task fixpix can select the type of pixel replacement to use for each mask value. So one can chose, for example, line interpolation for the linterp values and the eqinterp values, and column interpolation for the cinterp values.

In addition to this task, pixel mask images may be made in a variety of ways. Any task which produces and modifies image values may be used. Some useful tasks are imexpr, imreplace, imcopy, text2mask and mkpattern. If a new image is specified with an explicit ".pl" extension then the pixel mask format is produced.

Examples

1. Two flat fields of exposures 1 second and 3 seconds are taken, overscan and zero corrected, and trimmed. These are then used to generate a CCD mask.

cl> imarith flat1 / flat2 ratio
cl> ccdmask ratio mask

Revisions

CCDMASK V2.11: This task is new.