Spatial-dictionary for collaborative representation classification of hyperspectral images

In this paper, we propose a spatial-dictionary (SD) for collaborative representation classification (SCRC) of hyperspectral images. The proposed method consists of four main steps. First, we extract spatial features using 2-D Gabor filters and stack them with spectral features. Second, the SD is constructed by incorporating the spatial information of sparse vectors into the dictionary optimization process. Third, a multiple-mapping kernel is exploited to further integrate spatial information into the CRC framework. Lastly, the test samples are allocated with the class labels. Experimental results obtained on two hyperspectral datasets demonstrate that the proposed SCRC method can yield higher classification accuracy with much lower computational cost when compared to other traditional classifiers.