Class dependent compressive-projection principal component analysis for hyperspectral image reconstruction

Random projections have been demonstrated to be an efficient dimensionality reduction technique for Hyperspectral Imagery (HSI). Compressive-Projection Principal Component Analysis (CPPCA) is an efficient receiver-side reconstruction technique that recovers HSI data from encore-side random projections. In this paper, after receiving random projections from the encoder, we utilize a relatively small amount of training (ground-truth) data to partition the image into several subsets (such that each subset represents a unique class/object) in the projected domain, and then employ the CPPCA reconstruction algorithm independently to every group. It is expected that such a class-dependent reconstruction of HSI data will be more reliable, since it is based on statistics that are representative of the dominant mixtures in the scene. Experimental results with HSI datasets reveal that the proposed method is superior in performance compared to traditional CPPCA.