A new coefficient estimation method when using PCA for spectral super-resolution

Hyperspectral imaging can provide non-destructive measurement of different materials in many research areas such as agriculture, medical, food processing, and mineralogy. However, hyperspectral imaging also has its limitations, including low-spatial resolution and high cost both in equipment and computation. Meanwhile, RGB imaging is usually low cost and with a higher spatial resolution. In this paper, we introduce a new coefficient estimation method when using a PCA-based dictionary learning method to recover spectral reflectance from RGB images. Different from previous PCA based methods, which invert the dictionary matrix directly, our method provides a more accurate way to estimate the coefficients, while keeping the estimated coefficients in the range given by the PCA variance, and ensure the recovered spectral reflectance maps to the original RGB values.