Hyperspectral Image Visualization Using Band Selection

Several simple but efficient hyperspectral image display approaches are proposed to use selected bands for Red-Green-Blue (RGB) color composite construction, where visualization-oriented spectral segmentation and integration are developed. A series of band selection algorithms, including minimum estimated abundance covariance (MEAC) and linear prediction (LP), are implemented and compared. The resulting color displays are evaluated in terms of class separability using a statistical classifier, and perceptual color distance. Experimental results demonstrate that the color composite displays using MEAC and LP-selected bands can outperform other band selection methods with low computational cost, and their performance is also better than those of one-bit transform (1BT) and principal component analysis (PCA)-based hyperspectral visualization methods in the literature.