ANISOTROPICALLY FOVEATED NONLOCAL WEIGHTS FOR JOINT SPARSE REPRESENTATION-BASED HYPERSPECTRAL CLASSIFICATION

Joint sparse representation has yielded significant advances in hyperspectral classification due to its ability to incorporate spatial information of neighboring pixels. However, challenges remain for exploring the interpixel correlation. In this paper, we propose anisotropically foveated nonlocal weights for joint sparse representation-based classification of hyperspectral image (HSI). To this end, two major aspects are involved: 1) different weights, which are determined by anisotropically foveated similarity, are assigned to different neighborhoods around the central test pixel. Anisotropic foveation operators involved in this step can mimic the non uniformity (i.e. center is sharp while periphery is blurred) of human visual system (HVS). 2) simultaneous orthogonal matching pursuit (SOMP) is utilized to obtain the coefficient matrix in joint sparse representation-based classifier (JSRC). Experiments conducted on the benchmark Indian Pines data demonstrate the promising performance of our proposed method.