Back to articles
Volume: 28 | Article ID: art00043
Reconstructing Spectra from RGB Images by Relative Error Least-Squares Regression
  DOI :  10.2352/issn.2169-2629.2020.28.42  Published OnlineNovember 2020

Spectral reconstruction (SR) algorithms attempt to map RGB- to hyperspectral-images. Classically, simple pixel-based regression is used to solve for this SR mapping and more recently patch-based Deep Neural Networks (DNN) are considered (with a modest performance increment). For either method, the 'training' process typically minimizes a Mean-Squared-Error (MSE) loss. Curiously, in recent research, SR algorithms are evaluated and ranked based on a relative percentage error, so-called MeanRelative-Absolute Error (MRAE), which behaves very differently from the MSE loss function. The most recent DNN approaches - perhaps unsurprisingly - directly optimize for this new MRAE error in training so as to match this new evaluation criteria.<br/> In this paper, we show how we can also reformulate pixelbased regression methods so that they too optimize a relative spectral error. Our Relative Error Least-Squares (RELS) approach minimizes an error that is similar to MRAE. Experiments demonstrate that regression models based on RELS deliver better spectral recovery, with up to a 10% increment in mean performance and a 20% improvement in worst-case performance depending on the method.

Subject Areas :
Views 23
Downloads 10
 articleview.views 23
 articleview.downloads 10
  Cite this article 

Yi-Tun Lin, Graham D. Finlayson, "Reconstructing Spectra from RGB Images by Relative Error Least-Squares Regressionin Proc. IS&T 28th Color and Imaging Conf.,  2020,  pp 264 - 269,

 Copy citation
  Copyright statement 
Copyright © Society for Imaging Science and Technology 2020
Color and Imaging Conference
color imaging conf
Society for Imaging Science and Technology