Rule Images in Supervised Classification

A rule image is a greyscale image that shows intermediate classification results for a selected class. The pixel values represent different metrics, which are described below.

With the following classification tools, ENVI creates one rule image where each band represents a rule image for each class:

• Mahalanobis Distance classification: Pixel values represent the distance from the class mean.
• Maximum Likelihood classification: Pixel values contain a maximum likelihood discriminant function with a modified Chi Squared probability distribution. Higher rule image values indicate higher probabilities.
• Minimum Distance classification: Pixel values represent the Euclidean distance from the class mean.
• Spectral Angle Mapper (SAM) classification: Pixel values represent the spectral angle in radians from the reference spectrum for each class. Lower spectral angles represent better matches to the endmember spectra.

With the following tools, ENVI creates multiple rule images, one for each class. For example, if a classification involves 10 classes, then 10 rule images will be created. To create rule images for these classification methods, toggle the Output Rule Images? option to Yes.

• Binary Encoding classification: Pixel values represent the percentage (0-100%) of bands that matched that class.
• Parallelepiped classification: Pixel values range from 0 to n (where n is the number of bands). They represent the number of bands that satisfied the parallelepiped criteria.
• Spectral Information Divergence (SID) classification: One rule image is created for each endmember. Pixel values represent the SID value (the output of the equation that defines SID for a pair of spectral vectors). Lower spectral divergence measures represent better matches to the endmember spectra.
• Support Vector Machine (SVM): Pixel values represent the percentage (0-100%) of bands that matched that class.