Disadvantages of the method

  • One needs several pixels for each class to be able to compute higher order moments reliably.

  • The solution is computed by the least square error estimator and it is therefore prone to errors introduced by individual measurements that may be severely wrong (outliers). Bosdogianni, Petrou and Kittler proposed an alternative robust estimator (Mixed Pixel Classification with Robust Statistics, IEEE Transactions on Geoscience and Remote Sensing, vol 35(3), pp 551-559,1997).
  Back to "Practical Details"
  Back to "Dealing with sets of mixed pixels"

Learn more on robust spectral unmixing