MOPEX Online Manual

Dual Outlier Detection

Figure 1: Cartoon of dual outlier detection.

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_DUAL_OUTLIER_FOR_RMASK trigger in the namelist and set the RMask_Fatal_BitPattern to use bit 2. Note that this is not the same as setting it to a value of 2. See Appendix 2: Fatal Bit Patterns for more information.

Spatial detection

Module Detect Outlier applied to background subtracted images finds all the pixels with values greater than a specified number of standard deviations. The input parameters used here are Detection Max Area, Detection Min Area, and Detection Threshold. The product of this step is called a detection map. Each contiguous cluster is assigned a number and the values of all the pixels in the cluster are set to this number. See the document entitled Image Segmentation for a detailed description of the module and the processing parameters.

Detection map interpolation

The processing is performed by module Mosaic Projection. The process is different from image interpolation. The non-zero input pixels set the values of the output pixels they overlap without any weighting. The process is ambiguous if two clusters are close, and there are output pixels that overlap input pixels from both clusters. The product of this step is interpolated detection maps (see Figure 2).

Figure 2: A sample interpolated detection map. Clusters 1,2, and 5 are point sources, clusters 3 and 4 are radhits. Other pixels are set either to 0 if they overlap non-outlier pixels in the input detection map, or to NaN (not a number) if they overlap no pixels in the input detection map.

It is performed by module Mosaic Dual Outlier. The input parameters that are used are Max Outlier Image and Max Outlier Fraction. Interpolated detection maps are stacked up. For each stack the number of spatial outlier pixels is found. If the fraction of outliers in the stack is smaller or equal to Max Outlier Fraction and the number of spatial outliers in the stack is smaller or equal to Max Outlier Image, then these pixels are now classified as dual (spatial and temporal) outliers, and the sign of the dual outlier pixels is changed (see Figure 3). For example, in Figure 3, suppose Max Outlier Fraction = 0.5 and Max Outlier Image = 3. Stack A is the perfect case when only one pixel in the stack is a spatial outlier (value of -12). Its sign is changed from positive to negative. Stack B has 4 pixels that are spatial outliers (1, 12, 5, 8). The fraction of spatial outliers is 0.4 < Max Outlier Fraction, but 4 > Max Outlier Image, so since the stack does not satisfy both conditions, their signs are unchanged and remain positive. In Stack C the fraction of spatial outlier is 0.6 > Max Outlier Fraction, so the signs remain unchanged. Stack D is the perfect case of a pixel in a point source detected in all overlapping images. The fraction of spatial outliers is 1. The product of this step are dual detection maps (see Figure 4).

Figure 3: Suppose Max Outlier Fraction = 0.5 and Max Outlier Image = 3. Stack A is the perfect case when only one pixel in the stack is a spatial outlier (value of -12). Its sign is changed from positive to negative. Stack B has 4 pixels that are spatial outliers (1, 12, 5, 8). The fraction of spatial outliers is 0.4 < Max Outlier Fraction, but 4 > Max Outlier Image, so since the stack does not satisfy both conditions, their signs are unchanged and remain positive. In Stack C the fraction of spatial outlier is 0.6 > Max Outlier Fraction, so the signs remain unchanged. Stack D is the perfect case of a pixel in a point source detected in all overlapping images. The fraction of spatial outliers is 1.

Figure 4: As a result of temporal detection, the sign of the pixels of clusters 3 and 4 has been changed. Also some of the pixels of cluster 2 have their sign changed. Cluster 5 is probably a moving source - the spatial detection suggests that it is real, but the temporal detection shows that it is no longer at the same coordinates in subsequent frames. One of the pixels in cluster 5 has its sign unchanged. This could happen if an overlapping frame has a radhit affecting the corresponding pixel.

The processing is done by module Level. The dual detection map is processed in order to eliminate detection of the outskirts of legitimate point sources as dual outliers. If a dual outlier belongs to a cluster where the majority of pixels are not dual outliers the chances are the pixel has been wrongly marked because it is on the edge of the point source.

The sign of wrongly marked pixels is flipped based on the namelist parameter Threshold Ratio. The following is done. If the number of negative pixels N- in a cluster with N pixels is smaller than the threshold

N-/N < Threshold Ratio

then their signs are flipped. If the number of positive pixels N+ in a cluster with N pixels is smaller than the threshold

N+/N < Threshold Ratio

then their signs are flipped. If Threshold Ratio = 0.5, which is the default, the pixels within each cluster will have the same sign.

The products of this step are the corrected dual detection maps (see Figure 5).

Figure 5: Suppose Threshold Ratio = 0.5. In Figure 4, cluster 2 has outlier fraction of 0.25, which is less than Threshold Ratio. The corrected dual map has all the pixels in cluster 2 set positive. Cluster 5 in Figure 4 has an outlier fraction of 0.9, which is greater than Threshold Ratio. The single positive pixel in cluster 5 has been set negative.