Spectral unmixing with Vertex Component Analysis

In this paper, Vertex Component Analysis based spectral unmixing is proposed as a preprocessing stage for hyperspectral image compression. The two approaches generally used in hyperspectral image compression are to compress it as a 3D cube or apply two stages of data reduction (spectral and spatial). In the first method, the hyperspectral image is considered as a cube and 3-D compression is applied which is directly extended from 2-D compression. In the second approach, compression is done spectrally and spatially. The second method of two stage reduction is found to be more efficient than the first method. VCA based spectral unmixing is used for mitigating spectral redundancy. From the experimental results, it is inferred that VCA based preprocessing is efficient than PPI based preprocessing. The quality metrics used for performance assessment are Maximum Absolute Error, Mean Absolute error, Mean Square Error and Relative Root Mean Square Error.