Research on Spectral Image Reconstruction Based on Nonlinear Spectral Dictionary Learning From Single RGB Image

Chu ZUO; De-hong XIE; Xiao-xia WAN

doi:10.3964/j.issn.1000-0593(2022)07-2092-09

Journals >Spectroscopy and Spectral Analysis >Volume 42 >Issue 7 >Page 2092 > Article

Spectroscopy and Spectral Analysis
Vol. 42, Issue 7, 2092 (2022)

Research on Spectral Image Reconstruction Based on Nonlinear Spectral Dictionary Learning From Single RGB Image

Chu ZUO^1、1;, De-hong XIE^{2、2; *;}, and Xiao-xia WAN^3、3;

Author Affiliations

¹1. School of Light Industry and Food, Nanjing Forestry University, Nanjing 210037, China

²2. College of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, China

³3. Hubei Province Engineering Technical Center for Digitization and Virtual Roprodcuction of Color Information of Culture Relics, Wuhan University, Wuhan 430079, China

show less

DOI: 10.3964/j.issn.1000-0593(2022)07-2092-09 Cite this Article

Chu ZUO, De-hong XIE, Xiao-xia WAN. Research on Spectral Image Reconstruction Based on Nonlinear Spectral Dictionary Learning From Single RGB Image[J]. Spectroscopy and Spectral Analysis, 2022, 42(7): 2092 Copy Citation Text

show less

Fig. 1. Overview of our proposed spectral reconstruction method

Download full size | View in the Article

The distribution of two data sets in color space from the spectral dataset for training the spectral dictionary(a): Munsell; (b): Pantone+Munsell

Fig. 2. The distribution of two data sets in color space from the spectral dataset for training the spectral dictionary
(a): Munsell; (b): Pantone+Munsell

Download full size | View in the Article

Fig. 3. RGB images under different illuminats and residual error map between their reconstructed spectral images and their original captured spectral images, the training dataset of the reconstructed spectral images is Munsell

Download full size | View in the Article

Fig. 4. RGB images under different illuminants and residual error map between their reconstructed spectral images and their original captured spectral images, the training dataset of the reconstructed spectral images is Munsell+Pantone

Download full size | View in the Article

方法	Munsell		Munsell+Pantone
方法	平均值	标准偏差	平均值	标准偏差
PCA	0.036 4	0.021 1	0.037 3	0.021 8
NLPCA	0.017 4	0.010 7	0.016 9	0.010 0
改进NLPCA	0.015 6	0.009 5	0.014 5	0.009 5

Table 1. RMSE results of reconstructed spectral reflectance for different train sets

View in the Article

光源	重建方法	Munsell+Pantone(训练) (平均值±标准偏差值)		Munsell(训练) (平均值±标准偏差值)
光源	重建方法	CAVE	UEA	CAVE	UEA
	SR_SR^[14]	0.254 9±0.053 0	0.354 3±0.100 0	0.304 9±0.052 0	0.343 7±0.073 6
A	NLPCA_SR^[23]	0.411 6±0.253 1	0.422 0±0.231 2	0.378 1±0.131 9	0.385 1±0.097 2
	本文方法	0.202 2±0.067 0	0.253 4±0.087 8	0.207 6±0.061 9	0.271 1±0.066 4
	SR_SR^[14]	0.254 9±0.053 0	0.354 3±0.100 0	0.304 9±0.052 0	0.343 7±0.073 6
D65	NLPCA_SR^[23]	0.411 5±0.103 0	0.414 6±0.114 7	0.368 9±0.095 6	0.375 8±0.092 8
	本文方法	0.202 2±0.069 7	0.237 4±0.087 6	0.217 6±0.060 7	0.293 6±0.088 8
	SR_SR^[14]	0.254 9±0.053 0	0.354 3±0.100 0	0.304 9±0.052 0	0.343 7±0.073 6
F2	NLPCA_SR^[23]	0.235 6±0.076 6	0.281 3±0.104 8	0.440 4±0.236 7	0.360 7±0.168 1
	本文方法	0.232 7±0.068 8	0.275 5±0.078 8	0.263 1±0.077 9	0.320 1±0.105 9
	SR_SR^[14]	0.254 9±0.053 0	0.354 3±0.100 0	0.304 9±0.052 0	0.343 7±0.073 6
Avg.	NLPCA_SR^[23]	0.352 9±0.144 2	0.372 6±0.150 2	0.395 8±0.154 7	0.373 9±0.119 4
	本文方法	0.212 4±0.068 5	0.255 4±0.084 7	0.229 4±0.066 8	0.294 9±0.087 0

Table 2. RMSE results of the reconstructed spectral images from their RGB images under different illuminants

Download Citation

Save the article for my favorites

Paper Information