Outlier Detection

Processing multiple datasets together allows for the identification of bad pixels or cosmic-rays that remain in each of the input images, many times at levels which were ambiguous for detection during ramp fitting. The outlier detection step implements the following algorithm to identify and flag any remaining cosmic-rays or other artifacts left over from previous calibrations:

  • build a stack of input data
    • all inputs will need to have the same WCS since outlier detection assumes the same flux for each point on the sky, and variations from one image to the next would indicate a problem with the detector during readout of that pixel
    • if needed, each input will be resampled to a common output WCS
  • create a median image from the stack of input data
    • this median operation will ignore any input pixels which have a weight which is too low (<70% max weight)
  • create “blotted” data from the median image to exactly match each original input dataset
  • perform a statistical comparison (pixel-by-pixel) between the median,blotted data with the original input data to look for pixels with values that are different from the mean value by more than some specified sigma based on the noise model
    • the noise model used relies on the error array computed by previous calibration steps based on the readnoise and calibration errors
  • flag the DQ array for the input data for any pixel (or affected neighboring pixels) identified as a statistical outlier

The outlier detection step serves as a single interface to apply this general process to any JWST data, with specific variations of this algorithm for each type of data. Sub-classes of the outlier detection algorithm have been developed specifically for

  • Imaging data
  • IFU spectroscopic data
  • TSO data
  • coronagraphic data
  • spectroscopic data

This allows the outlier_detection step to be tuned to the variations in each type of JWST data.

Outlier Detection Code API

jwst.outlier_detection Package

Classes

OutlierDetectionStep([name, parent, …]) Flag outlier bad pixels and cosmic rays in DQ array of each input image.
OutlierDetectionScaledStep([name, parent, …]) Flag outlier bad pixels and cosmic rays in DQ array of each input image.
OutlierDetectionStackStep([name, parent, …]) Class definition for stacked outlier detection.

Class Inheritance Diagram

Inheritance diagram of jwst.outlier_detection.outlier_detection_step.OutlierDetectionStep, jwst.outlier_detection.outlier_detection_scaled_step.OutlierDetectionScaledStep, jwst.outlier_detection.outlier_detection_stack_step.OutlierDetectionStackStep