ERSC 3P07 Lecture 8: Accuracy Assessment - Microwave Remote Sensing
Document Summary
Classification results often variable due to spectral variability of the data. Often, a few pixels from an otherwise homogeneous area are classified as something else. This technique is generally called spatial filtering or smoothing: Quantitative assessments of image classification accuracy are not always possible. Example: however, quantitative assessments should be conducted whenever possible. Accuracy defines correctness (measures agreement between a standard and an image classification of unknown quality). Precision defines detail where increasing detail may lead to increasing opportunity for error. Measures agreement between a standard assumed to be correct and an image classification of unknown quality; if classification corresponds closely with the standard, it is said to be. Several indices are used to measure the degree of correspondence including: Aka: error matrix (essentially, it is a contingency table) A cross-tabulation of classified pixels compared to pixels of known value m x m matrix where m = # of classes.