Gabor-Filtering-Based Nearest Regularized Subspace for Hyperspectral Image Classification

Abstract

By coupling the nearest-subspace classification with a distance-weighted Tikhonov regularization, nearest regularized subspace (NRS) was recently developed for hyperspectral image classification. However, the NRS was originally designed to be a pixel-wise classifier which considers the spectral signature only while ignoring the spatial information at neighboring locations. Gabor features have currently been successfully applied for hyperspectral image analysis due to the ability to represent useful spatial information. In this paper, we mainly exploit the benefits of using spatial features extracted from a simple Gabor filter for the NRS classifier. The proposed Gabor-filtering-based classifier has been validated on several real hyperspectral datasets. Experimental results demonstrate that the proposed method significantly increases the classification accuracy compared to conventional pixel-wise classifiers as well as Gabor-filtering-based support vector machine and sparse-representation-based classification.