Hyperspectral image classification using multiple features and nearest regularized subspace

Gabor features have been proved to be effective for the recently-proposed nearest regularized subspace (NRS) classifier. In this paper, we further investigate a residual fusion based strategy with multiple features and NRS. Multiple features include local binary patterns (LBP), Gabor features and the original spectral signatures. In the proposed classification framework, each type of feature is first coupled with the NRS classifier, obtaining the output of residuals. And then, all the residuals are added together and the label of the test pixel is determined accorDing to the minimum residual. The motivation of this work is due to that different features represent the test pixel from different perspectives and the fusion in the residual domain is able to enhance the discriminative ability, especially for small-sample-size situations. Experimental results of several hyperspectral image datasets demonstrate that the proposed residual-based fusion strategy is superior to the traditional NRS and Gabor-NRS.