Using principal component analysis for enhancement of multispectral infrared telescope images

The performance of Principal Component Analysis (PCA) on multispectral astronomical data sets obtained from the Spatial Infrared Imaging Telescope (SPIRIT III) aboard the Midcourse Space Experiment satellite was investigated. Covariance and correlation methods of PCA were studied. The first principal component demonstrated the success of PCA as a multispectral image fusion technique. Higher order principal components provided feature discrimination of stars with different spectral classification and of molecular clouds with varying physical and chemical properties. The intrinsic dimensionality was two or three components, depending on the scene. PCA was found to be a channel capacity enhancement technique by using the intrinsic dimensionality to reduce the size of the data. Additional data compression was found in certain principal components containing near-zero eigenvector elements. Covariance PCA was found to have better performance because of more aggressive accounting of the total variance and aesthetics of images produced for human interpretation.