Fig 6. Classification accuracy of our method when trained with features extracted from varying number of selected bands.

Various numbers of bands are selected using a method based on similarity based criteria, as described in [7]. (a) shows the results of selecting varying number of bands for each texture descriptor separately along with our ensemble approach (b) shows the results obtained for 2-class and 4-class classification. 2-class classification consists of normal and CA class while 4-class classification has all the original classes.