STScI Logo

crmedian noao.imred.crutil



crmedian -- detect, fix, and flag cosmic rays using median filtering


crmedian input output


Input image in which to detect cosmic rays.
Output image in which cosmic rays are replaced by the median value. If no output image name is given then no output image will be created.
crmask = ""
Output cosmic ray mask. Detected cosmic rays (and other deviant pixels) are identified in the mask with values of one and good pixels with a values of zero. If no output cosmic ray mask is given then no mask will be created.
median = ""
Output median filtered image. If no image name is given then no output will be created.
sigma = ""
Output sigma image. If no image name is given then no output will be created.
residual = ""
Output residual image. This is the input image minus the median filtered image divided by the sigma image. Thresholds in this image determine the cosmic rays detected. If no image name is given then no output will be created.
var= 0., var1 = 0., var2 = 0.
Variance coefficients for the variance model. The variance model is

    variance = var0 + var1 * data + var2 * data^2

where data is the maximum of zero and median pixel value and the variance is in data numbers. All the coefficients must be positive or zero. If they are all zero then empirical data sigmas are estimated by a percentile method in boxes of size given by ncsig and nlsig .

lsigma = 10, hsigma = 3
Positive sigma factors to use for selecting pixels below and above the median level based on the local percentile sigma. Cosmic rays will appear above the median level.
ncmed = 5, nlmed = 5
The column and line size of a moving median rectangle used to estimate the uncontaminated local image.
ncsig = 25, nlsig = 25
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.


Crmedian detects cosmic rays from pixels deviating by a specified statistical amount from the median at each pixel. It outputs and set of the following: a copy of the input image with cosmic rays replaced by the median value, a cosmic ray mask identifying the cosmic rays, the median filtered image, a sigma image where each pixel has the estimated sigma, and the residual image used in detecting the cosmic rays.

The residual image is computed by subtracting a median filtered version of the input data from the unfiltered input data and dividing by an estimate of the pixel sigmas. The median filter box size is given by the ncmed and nlmed parameters. If a name for the median image is specified the median filtered image will be output. The variance at each pixel is determined either from a variance model or empirically. If a name for the sigma image is specified then the sigma values (the square root of the variance) will be output. If a name for the residual image is given then the residual image will be output.

The empirical variance model is given by the formula

    variance = var0 + var1 * data + var2 * data^2

where data is the maximum of zero and median pixel value and the variance is in data numbers. This model can be related to common detector parameters. For CCDs var0 is the readout noise expressed as a variance in data numbers and var1 is the inverse gain (DN/electrons). The second order coefficient has the interpretation of flat field introduced variance.

If all the coefficients are zero then an empirical sigma is estimated as follows. The input image is divided into blocks of size ncsig and nlsig . The pixel values in a block are sorted and the pixel values nearest the 15.9 and 84.1 percentiles are selected. These are the one sigma points in a Gaussian distribution. The sigma estimate is the difference of these two values divided by two. This algorithm is used to avoid contamination of the sigma estimate 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 points. The sigma estimate for a pixel is the sigma from the nearest block. A moving box is not used for efficiency.

The residual image is divided by the sigma estimate at each pixel. Cosmic rays are identified by finding those pixels in the residual image which have values greater than hsigma and bad pixels with values below lsigma are also identified.

If an output image name is specified then the output image is produced as a copy of the input image but with the identified cosmic ray pixels replaced by the median value. If an output cosmic ray mask is specified a cosmic ray mask will be produced with values of zero for good pixels and one for bad pixels. The cosmic ray mask is used to display the cosmic ray positions found and the cosmic rays can be replaced by interpolation (as opposed to the median value) using the task crfix .

The crmedian detections are very simple and do not take into account real structure with scales of a pixel. Thus this may clip the cores of stars and narrow nebular features in the data. More sophisticated algorithms are found in cosmicrays , craverage , and crnebula . The median, sigma, and residual images are available as output to evaluate the various aspects of the algorithm.


This example illustrates using the crmedian task to give a cosmic ray removed image and examining the results with an image display. The image is a CCD image with a readout noise of 5 electrons and a gain of 3 electrons per data number. This implies variance model coefficients of

    var0 = (5/3)^2 = 2.78
    var1 = 1/3 = 0.34

    cl> display obj001 1                  # Display in first frame
    cl> # Determine output image, cosmic ray mask, and residual image
    cl> crmedian obj001 crobj001 crmask=mask001 resid=res001\
    >>> var0=2.78 var1=0.34
    cl> display crobj001 2                # Display final image
    cl> display mask001 3 zs- zr- z1=-1 z2=2 # Display mask
    cl> display res001 4 zs- zr- z1=-5 z2=5  # Display residuals

By looking at the residual image the sigma clippig threshold can be adjusted and the noise parameters can be tweaked to minimize clipping of real extended structure.


cosmicrays, craverage, crnebula, median, crfix, crgrow

Source Code · Search Form · STSDAS