PICCRS: Optimally Combine Frames with Outlier Rejection



NOTE!!! PICCRS can be a complicated program. Make sure to understand and check your results; we offer no guarantees. If your images are not well-registered, PICCRS won't work!

PICCRS is a program originally written by Ed Groth at Princeton for the statistical removal of cosmic rays from HST WFPC data. It is useful, however, as an optimal combiner of frames with rejection of outlying pixels. PICCRS works by going through each pixel in a stack of images. For each pixel, the value in each image is compared with the mean determined from all of the other images in the stack. Using the known noise properties of the detector, the pixel in question will be rejected if it falls outside some specified criterion. This continues through the entire stack. Finally, the weighted mean of all pixels not flagged is taken and placed in the output image.

The input images are specified in a file given with the LIST= file, with format exactly like that of BIGMEDIAN. A mean and bias value can be specified in this input file. There is an optional fourth parameter which can be specified, unlike BIGMEDIAN, which is an output file name for each input file; if this name is given, the output file will be identical to the input file but all of the rejected pixels will be set to BLANK. The output file will be in WFPC format with a separate header and image file.

By default the input files should be in FITS format, but WFPC input files can be used if the WFPC keyword is specified.

IMPORTANT: To insure proper operation, the gain and readout noise must be correctly specified with the GAIN= and RN= keywords. The default values of GAIN=7.5 and RN=13 apply for WF/PC-1 data (NOT WFPC2). If desired, saturated pixels can be rejected using the BSAT=n keyword which specifies that all pixels larger than n will be considered bad on input. IMPORTANT: by default, all negative pixels are considered bad on input; you can change the lower boundary for bad pixels with the MIN=x keyword.

Since the original idea of PICCRS was to reject cosmic rays, the program is designed to allow a less stringent rejection criterion for neighbors of pixels which deviate by a lot, to allow for the fact that cosmic rays often take up several pixels, and that pixels next to the brightest cosmic ray pixels are more likely to be part of the cosmic ray hit than any random pixel in the image. Consequently, the algorithm goes through the image and rejects as "primary" cosmic rays those pixels above the threshold specified with the TP= keyword (default is 5 (sigma)), then neighbors of these pixels are tested with a lower threshold specified with the TN= (for adjacent neighbors) and TD= (for diagonal neighbors) keywords (defaults are 3 (sigma) for both of these).

The whole idea of PICCRS depends on the fact that the input images are all well registered; if this is not the case, then most of the objects, e.g. stars, will be rejected as deviant from the mean and your output image will be horribly incorrect. To account for very small errors in registration, the computed errors, against which the pixel deviations are tested for rejection, includes a "fudge" factor to increase the expected variance for an increased signal level; this helps prevent bright pixels in stars with slight misregistration from being rejected as cosmic rays. Note that if such pixels are rejected, the output image will not be photometrically correct. The variance computed by PICCRS is given by:

   variance(DN) = (rn/gain)**2 + (DN/GAIN) + (F*DN)**2
where the first two terms are strictly correct and the last term is the fudge factor with variable parameter F. The default for F is 0.05, but this can be modified with the F= keyword; to turn off the fudge factor, set F to 0, but make sure your (real) bright objects aren't getting clipped.

MEDIAN keyword.

BLANK= keyword

NEG keyword.