Distance evaluation on spectral reconstruction

Spectral distance and similarity measures inherently bias optimization toward spectral shape or total radiance. In this work, we analyze how such biases affect reconstruction outcomes by comparing several commonly used metrics, including SAM, SGA, SCA, SID, the Euclidean distance, and the Kullback-Leibler pseudo-divergence (KLPD). Distances are evaluated through a spectral reconstruction task based on distance-driven optimization. By exploiting the KLPD decomposition into shape and radiance terms, our results provide a finer interpretation of reconstruction errors and reveal how each metric biases the solution space. This study highlights the relevance of the KLPD for analyzing and comparing spectral distance measures in reconstruction problems.