MOPEX Online Manual
| Contents | ~ | Previous: Single Frame Outlier Detection | ~ | Next: Dual Outlier Detection |
MultiFrame Temporal Outlier Detection
Figure 1: Cartoon of temporal outlier detection.
The second outlier detection method carries out temporal filtering of the interpolated images. It is performed by the module Mosaic Outlier. The input parameters used for this module are Bottom Threshold, Top Threshold,, and Min Pix Num.
Important: Using this module does not mean that MOPEX will automatically use the results for outlier detection. In order to use the results from this module, you must include the USE_OUTLIER_FOR_RMASK trigger in the namelist and set the RMask_Fatal_BitPattern to use bit 1. Note that this is not the same as setting it to a value of 1. See Appendix 2: Fatal Bit Patterns for more information.
The input images are interpolated onto a common grid, such that the pixels in the interpolated images are perfectly lined up and create a stack of pixels. Each spatial location, i.e. a pixel in the common grid, is filtered in the time domain and outliers are found (see Figure 2).
If pixel value Ik for pixel k in the stack satisfies either of the following conditions:
Ik < M - Bottom_Threshold x σ
Ik > M + Top_Threshold x σ
pixel k is declared an outlier. M and σ are estimates of the mean and scatter of the pixel values in the stack. If pixel uncertainty images are provided σ above is set to the greater of two values: the scatter and the smallest pixel uncertainty in the stack. The scatter can not be reliably estimated with only a handful of samples. If the stack size is less than Min Pix Num, then σ is set to the smallest pixel uncertainty in the stack.
Two ways to estimate the mean M and scatter σ of the pixel values in the stack are implemented. It is controlled by the namelist variable THRESH_OPTION.
THRESH_OPTION = 1 is the more robust and highly recommended way of sigma estimation. Under this option,
M = median(Ik), σ=MAD(Ik) = median(|Ik-median(Ii)|)/0.6745
where MAD stands for the median absolute deviation. The factor of 0.6745 here represents the inverse of the third quartile of the normal distribution. The division is necessary because the numerator alone tends to underestimate the standard deviation, so dividing by this value makes the MAD more accurate.
THRESH_OPTION = 2 is less robust and based on the minimum difference between the values of two consecutive pixels in the stack. We strongly recommend that users do not use this option.
Ok = (Ik-M)/σ
The thresholds - Bottom Threshold and Top Threshold - can be set to 0, which is the default. In this case the decision of declaring a particular pixel an outlier is put off until running Mosaic RMask. The advantage of doing it this way is that the user can experiment with different thresholds for outliers without having to rerun mosaic_outlier module.
The main limitation of temporal outlier rejection is the cases of shallow coverage.
| Contents | ~ | Previous: Single Frame Outlier Detection | ~ | Next: Dual Outlier Detection |
