Improving RGB illuminant estimation exploiting spectral average radiance

Ilaria Erba; Marco Buzzelli; Jean-Baptiste Thomas; Jean-Baptiste Thomas; Jon Yngve Hardeberg; Raimondo Schettini

doi:10.1364/JOSAA.510159

Journal of the Optical Society of America A
Vol. 41,
Issue 3,
pp. 516-526
(2024)
•https://doi.org/10.1364/JOSAA.510159

Improving RGB illuminant estimation exploiting spectral average radiance

Ilaria Erba, Marco Buzzelli, Jean-Baptiste Thomas, Jon Yngve Hardeberg, and Raimondo Schettini

Not Accessible

Your library or personal account may give you access

Get PDF
Email
Share
Get Citation
Copy Citation Text
Ilaria Erba, Marco Buzzelli, Jean-Baptiste Thomas, Jon Yngve Hardeberg, and Raimondo Schettini, "Improving RGB illuminant estimation exploiting spectral average radiance," J. Opt. Soc. Am. A 41, 516-526 (2024)

Export Citation
- BibTex
- Endnote (RIS)
- HTML
- Plain Text
Citation alert
Save article

Check for updates

Abstract

We introduce a method that enhances RGB color constancy accuracy by combining neural network and $k$-means clustering techniques. Our approach stands out from previous works because we combine multispectral and color information together to estimate illuminants. Furthermore, we investigate the combination of the illuminant estimation in the RGB color and in the spectral domains, as a strategy to provide a refined estimation in the RGB color domain. Our investigation can be divided into three main points: (1) identify the spatial resolution for sampling the input image in terms of RGB color and spectral information that brings the highest performance; (2) determine whether it is more effective to predict the illuminant in the spectral or in the RGB color domain, and finally, (3) assuming that the illuminant is in fact predicted in the spectral domain, investigate if it is better to have a loss function defined in the RGB color or spectral domain. Experimental results are carried out on NUS: a standard dataset of multispectral radiance images with an annotated spectral global illuminant. Among the several considered options, the best results are obtained with a model trained to predict the illuminant in the spectral domain using an RGB color loss function. In terms of comparison with the state of the art, this solution improves the recovery angular error metric by 66% compared to the best tested spectral method, and by 41% compared to the best tested RGB method.

Full Article | PDF Article

More Like This

RGB color constancy using multispectral pixel information

Ilaria Erba, Marco Buzzelli, and Raimondo Schettini
J. Opt. Soc. Am. A 41(2) 185-194 (2024)

Autoencoder-based training for multi-illuminant color constancy

Donik Vršnak, Ilija Domislović, Marko Subašić, and Sven Lončarić
J. Opt. Soc. Am. A 39(6) 1076-1084 (2022)

Illuminant estimation in multispectral imaging

Haris Ahmad Khan, Jean-Baptiste Thomas, Jon Yngve Hardeberg, and Olivier Laligant
J. Opt. Soc. Am. A 34(7) 1085-1098 (2017)

Previous Article Next Article

Supplementary Material (1)

Name	Description
Supplement 1	Supplemental document.

Data availability

The work and results presented in this paper are based on the original NUS dataset by Nguyen et al. [35].

35. R. M. Nguyen, D. K. Prasad, and M. S. Brown, “Training-based spectral reconstruction from a single rgb image,” in European Conference on Computer Vision (Springer, 2014), pp. 186–201.

Cited By

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Figures (10)

You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Tables (3)

You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Equations (2)

You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Method	Patch Size	Min	Mean	Median	ple95	Max
CA	4	0.03	1.92	0.68	9.37	26.85
	8	0.03	1.71	0.37	7.16	15.41
	16	0.03	1.93	0.18	7.52	16.96
	32	0.03	1.80	0.20	7.58	17.25
	64	0.03	1.81	0.55	6.72	15.91
	128	0.03	2.04	1.12	6.34	15.53
	256	0.06	1.98	1.09	6.36	15.96
	512	0.08	2.52	1.94	4.90	13.86
SATOS	4	0.03	2.34	0.32	9.44	39.72
	8	0.03	2.10	0.47	8.24	29.14
	16	0.03	2.29	0.21	12.16	17.26
	32	0.03	1.93	0.22	9.17	15.50
	64	0.03	1.97	0.31	10.20	15.52
	128	0.03	1.84	1.06	7.65	15.02
	256	0.03	2.30	1.65	6.81	16.74
	512	0.76	2.90	1.79	6.92	12.31
SATOC	4	0.03	1.90	0.04	10.07	45.25
	8	0.03	2.21	0.10	9.45	21.22
	16	0.03	1.91	0.13	8.20	17.39
	32	0.03	1.84	0.37	7.60	15.46
	64	0.03	1.67	0.73	6.56	14.07
	128	0.03	1.77	0.80	6.42	15.40
	256	0.03	1.80	0.83	7.09	16.26
	512	0.69	2.80	1.87	6.40	11.80

Method	Patch Size	Selection Cluster Accuracy	Centroid Recovery Error	Closest to Centroid Recovery Error
CA	256	64%	1.82	1.82
	128	56%	1.98	2.00
	64	76%	1.57	1.58
	32	72%	1.53	1.52
	16	64%	1.76	1.75
	8	88%	1.33	1.33
	4	80%	1.58	1.58
SATOC	256	76%	1.78	1.80
	128	56%	1.83	1.83
	64	68%	1.56	1.57
	32	72%	1.75	1.74
	16	80%	1.46	1.46
	8	76%	1.90	1.90
	4	80%	1.63	1.64
SATOS	256	44%	2.28	2.22
	128	48%	1.78	1.76
	64	60%	1.62	1.63
	32	64%	1.89	1.89
	16	64%	2.31	2.31
	8	68%	1.86	1.86
	4	64%	2.04	2.05

Method	Mean Recovery Angular Error	% Improvement Mean Recovery Error
Our baseline ( ${CA}_{8}$ )	1.33	\
Robles-Kelly et al. [26]	12.56	89%
Khan et al. [24]	3.96	66%
First Order Grey-Edge (GE1) [14]	5.46	76%
Second Order Grey-Edge (GE2) [14]	5.55	76%
General Grey World (GGW) [14]	3.67	64%
Grey World (GW) [14]	3.81	65%
Shades of Grey (SoG) [14]	3.84	65%
White Point (WP) [14]	4.81	72%
Grayness index (GI) [36]	2.41	45%
PCA [40]	3.12	58%
Quasi-unsupervised (QU) [41]	2.25	41%
SIIE [38]	6.19	79%
Adaptive surround modulation (ASM) [39]	3.69	64%

Method	Patch Size	Min	Mean	Median	ple95	Max
CA	4	0.03	1.92	0.68	9.37	26.85
	8	0.03	1.71	0.37	7.16	15.41
	16	0.03	1.93	0.18	7.52	16.96
	32	0.03	1.80	0.20	7.58	17.25
	64	0.03	1.81	0.55	6.72	15.91
	128	0.03	2.04	1.12	6.34	15.53
	256	0.06	1.98	1.09	6.36	15.96
	512	0.08	2.52	1.94	4.90	13.86
SATOS	4	0.03	2.34	0.32	9.44	39.72
	8	0.03	2.10	0.47	8.24	29.14
	16	0.03	2.29	0.21	12.16	17.26
	32	0.03	1.93	0.22	9.17	15.50
	64	0.03	1.97	0.31	10.20	15.52
	128	0.03	1.84	1.06	7.65	15.02
	256	0.03	2.30	1.65	6.81	16.74
	512	0.76	2.90	1.79	6.92	12.31
SATOC	4	0.03	1.90	0.04	10.07	45.25
	8	0.03	2.21	0.10	9.45	21.22
	16	0.03	1.91	0.13	8.20	17.39
	32	0.03	1.84	0.37	7.60	15.46
	64	0.03	1.67	0.73	6.56	14.07
	128	0.03	1.77	0.80	6.42	15.40
	256	0.03	1.80	0.83	7.09	16.26
	512	0.69	2.80	1.87	6.40	11.80

Method	Patch Size	Selection Cluster Accuracy	Centroid Recovery Error	Closest to Centroid Recovery Error
CA	256	64%	1.82	1.82
	128	56%	1.98	2.00
	64	76%	1.57	1.58
	32	72%	1.53	1.52
	16	64%	1.76	1.75
	8	88%	1.33	1.33
	4	80%	1.58	1.58
SATOC	256	76%	1.78	1.80
	128	56%	1.83	1.83
	64	68%	1.56	1.57
	32	72%	1.75	1.74
	16	80%	1.46	1.46
	8	76%	1.90	1.90
	4	80%	1.63	1.64
SATOS	256	44%	2.28	2.22
	128	48%	1.78	1.76
	64	60%	1.62	1.63
	32	64%	1.89	1.89
	16	64%	2.31	2.31
	8	68%	1.86	1.86
	4	64%	2.04	2.05

Abstract

Supplementary Material (1)

Data availability

Cited By

Figures (10)

Tables (3)

Equations (2)

Journal of the Optical Society of America A