What is the role of using 'Subspace Background' when performing Target Detection?

What is the role of using subspace background in Target Detection?

Many statistics-based spectral detection methods need to model the scene background with a covariance or correlation matrix. Subspace background is a technique used to calculate the background statistics by removing anomalous signatures from the background model. By better characterizing the background, spectral detection methods that use the subspace background technique achieve greater target to background separation and have great potential to improve the detection result.

The Subspace Background option is available for the following spectral tools: Matched Filter (MF), Adaptive Coherence Estimator (ACE) and Constrained Energy Minimization (CEM).

What is the rationale for setting the subspace threshold between 0.5 – 1?

The Background Threshold is used to specify what fraction of the anomalous image to use for calculating the subspace background statistics. For example, setting a background threshold to 0.9 uses 90% of the image pixels that have the lowest anomalous values to estimate the statistics of the background. The allowable range is 0.500 to 1.000 and the default is 0.900. If it is set to 1.000, the background statistics is calculated using the whole image.  The default value of 0.9 is a good value for most applications.  If the image has many anomalous pixels (>10%) which is not common, the user can decrease this value.

Does using background spectra always result in an increased number of target pixels?

It is not true that using background spectra always result in increased number of target pixels. As explained above, using subspace background often improves the estimation of the background statistics.

Review on 12/31/2013 MM