Optimized light source spectral power distribution for RGB camera based spectral reflectance recovery

Lixia Wang,Xiaoxia Wan,Aditya Sole,Jon Yngve Hardeberg

doi:10.1364/oe.425401

Abstract

The accuracy of recovered spectra from camera responses mainly depends on the spectral estimation algorithm used, the camera and filters selected, and the light source used to illuminate the object. We present and compare different light source spectrum optimization methods together with different spectral estimation algorithms applied to reflectance recovery. These optimization methods include the Monte Carlo (MC) method, particle swarm optimization (PSO) and multi-population genetic algorithm (MPGA). Optimized SPDs are compared with D65, D50 A and three LED light sources in simulation and reality. Results obtained show us that MPGA has superior performance, and optimized light source spectra along with better spectral estimation algorithm can provide a more accurate spectral reflectance estimation of an object surface. Meanwhile, it is found that camera spectral sensitivities weighted by optimized SPDs tend to be mutually orthogonal.

Highlights

Spectral reflectance property of a material tells how strongly the material reflects light incident on it and can be defined as the material’s ‘fingerprint’
It is clear that: the performance of multi-population genetic algorithm (MPGA) is significantly better than particle swarm optimization (PSO) for most spectral estimation algorithms except for radial basis function network (RBF), local-weighted linear regression model (LLR) and sequential weighted nonlinear regression (SWNR) in Munsell dataset; the superiority of MPGA over PSO is significant for pseudo-inverse estimation (PI), regularized least squares (RLS), LLR, SSR but insignificant for regularized local linear model (RLLM), RBF, kernel based model (Kernel), weighted nonlinear regression model (WNR), SWNR in SFU dataset; the differences between MPGA and PSO for all the spectral estimation algorithms except for LLR are significant in TC3.5 dataset; the superiority of MPGA and PSO over Monte Carlo (MC) is significant for almost every spectral estimation algorithm
We used MC, PSO and MPGA based optimization techniques along with PI, RLS [12], RLLM [13], Kernel [14], RBF [29], WNR [15], LLR [16], smoothest reflectance reconstruction (SRR) [28] and SWNR [17] estimation algorithms for optimizing a tunable light-emitting diodes (LEDs) light source spectral power distributions (SPDs) to improve the accuracy of estimating the spectral reflectance of an object surface

Summary

Introduction

Spectral reflectance property of a material tells how strongly the material reflects light incident on it and can be defined as the material’s ‘fingerprint’. The methods presented in [10,11,12,13,14,15,16,17] are indirect and less accurate as they obtain an estimation of the spectral reflectance from a reduced number of channel responses. Many methods have been proposed to estimate spectral information from camera responses These methods fall into three branches: traditional, machine learning, and deep learning methods. Heikkinen et al [25] and Eckhard et al [14] proposed different kernel-based regression models to recover reflectance and obtained a relatively high accuracy. Zhang et al [13] recovered reflectance by a linear combination of k reflectances from training set that have similar camera responses to target sample. Burns [28] found the smoothest reflectance in metamer set by using a hyperbolic tangent transformation to limit reflectance between 0 and 1, and recovered reflectance with a higher accuracy than Li and Luo

Methods

Results

Conclusion