Hyperspectral image segmentation using filter banks for texture augmentation

Abstract

This paper presents a method for appending texture information to existing hyperspectral data to increase classification accuracy. The features extracted for texture classification are based on the subbands of various configurations of the octave-band directional filter bank. This filter bank represents a computationally efficient alternative to other 2-D decompositions, and it is able to divide frequency space into equivalent and meaningful partitions. Results using different radial and angular resolutions are presented, and the different filter bank configurations are compared and discussed with respect to other decompositions.