On Statistical Band Selection for Image Visualization
Myriam Beauchemin, Ko Fung · Photogrammetric Engineering & Remote Sensing · 2001
Two often-cited statistical band selection methods for image visualization have been compared and shown not to agree when applied to the same data set. This discrepancy is mainly caused by the weight assigned to the individual band variance term in each index. An alternative method based on the minimization of redundant information between bands is presented. The new measure is robust against the existence of multicollinearity. A procedure to help determine an appropriate band subset size is also proposed.