Dimension reduction and entropy-based hyperspectral image visualization using hue, saturation, and brightness

Hyperspectral imaging (HSI) captures rich spectral data across hundreds of contiguous bands for diverse applications. Dimension reduction (DR) techniques are commonly used to map the first three reduced dimensions to the red, green, and blue channels for RGB visualization of HSI data. In this study, we propose a novel approach, HSBDR-H, which defines pixel colors by first mapping the two reduced dimensions to hue and saturation gradients and then calculating per-pixel brightness based on band entropy so that pixels with high intensities in informative bands appear brighter. HSBDR-H can be applied on top of any DR technique, improving image visualization while preserving low computational cost and ease of implementation. Across all tested methods, HSBDR-H consistently outperformed standard RGB mappings in image contrast, structural detail, and informativeness, especially on highly detailed urban datasets. These results suggest that HSBDR-H can complement existing DR-based visualization techniques and enhance the interpretation of complex hyperspectral data in practical applications. Tested in remote sensing applications involving urban and agricultural datasets, the method shows potential for broader use in other disciplines requiring high-dimensional data visualization.