Designing Quadcolor Cameras with Conventional RGB Channels to Improve the Accuracy of Spectral Reflectance and Chromaticity Estimation
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper investigates the use of quadcolor cameras with a fourth channel to improve the estimation of spectral reflectance and chromaticity. It is worth noting that quadcolor cameras use conventional RGB channels and a fourth channel to improve accuracy, this channel being constituted by a silicon sensor with optimized optical filter.
The paper also indicates that several methods have been proposed for the recovery of spectral reflectance, including basis spectra and Wiener estimation. The II-LUT method stands out for its independence in relation to irradiance, while traditional methods face problems of irradiance dependence, requiring specific training samples. It is worth noting that the II-LUT method, adopted by the research, presented higher accuracy compared to the base spectra method.
Finally, the paper analyzed the use of optical filters (band-pass and notch filters) used to improve the accuracy of the estimation of reflectance and chromaticity. The results showed that notch filters are more effective, effectively reducing reflectance and chromaticity errors.
Within the context presented, we consider that the paper is well structured in terms of theoretical contextualization, methodology and with interesting results. As a suggestion, we recommend the preparation of a flowchart of the methodology used for better understanding.
Author Response
Comments:
The paper investigates the use of quadcolor cameras with a fourth channel to improve the estimation of spectral reflectance and chromaticity. It is worth noting that quadcolor cameras use conventional RGB channels and a fourth channel to improve accuracy, this channel being constituted by a silicon sensor with optimized optical filter.
The paper also indicates that several methods have been proposed for the recovery of spectral reflectance, including basis spectra and Wiener estimation. The II-LUT method stands out for its independence in relation to irradiance, while traditional methods face problems of irradiance dependence, requiring specific training samples. It is worth noting that the II-LUT method, adopted by the research, presented higher accuracy compared to the base spectra method.
Finally, the paper analyzed the use of optical filters (band-pass and notch filters) used to improve the accuracy of the estimation of reflectance and chromaticity. The results showed that notch filters are more effective, effectively reducing reflectance and chromaticity errors.
Within the context presented, we consider that the paper is well structured in terms of theoretical contextualization, methodology and with interesting results. As a suggestion, we recommend the preparation of a flowchart of the methodology used for better understanding.
Response:
Thank you for taking the time to review our article and for giving us a high rating. We hope that our article will be useful to the scientific community. Based on your comments, we have added a new figure to show the flow chart of interpolating a test sample using the II-LUT method in Section 3.3, where the method is condensed into five steps.
Reviewer 2 Report
Comments and Suggestions for AuthorsThis phrasing is confusing: 'The same as Fig. 4, except ...'
Comments for author File: 
Comments.pdf
Author Response
Thank you for taking the time to review our article. We hope that our article will be useful to the scientific community.
Comment 1: In general, the structure of the manuscript could be improved as there are some confusing paragraphs. I do not consider it to be an accurate or concise description of your main results. If the focus were narrowed to spectral band rejection filters, where the best results are achieved, the 'Results and Discussion' section could be condensed. In the Methods section, I think the description of the methodology could be improved and condensed.
Response 1: If only spectral reflectance recovery is desired, the optimization cost function can be set to the mean ERef. However, color reproduction applications require low color differences. Therefore, we consider a cost function defined as Eq. (8), where ERef,μ and ΔE00,μ are the mean ERef and mean ΔE00, respectively; the Lagrange multiplier γ is a weighting factor that controls the contribution of color difference. The optimal design is a trade-off between reflectance accuracy and chromaticity accuracy, depending on the specific application requirements. Therefore, we present a number of optimal designs rather than a single best result.
We have added a new figure to show the flow chart of interpolating a test sample using the II-LUT method in Section 3.3, where the method is condensed into five steps.
Comment 2: In general, in introduction section, after of background and literature review, it ends describing the research purpose and methods. But not describing a summary of the main contributions of purpose. Them can be synthetized in conclusions.
Response 2: Based on your comment, we revised the paragraph from lines 86-91 and added a new paragraph at the end of the Conclusions section to make our contribution more focused.
Comment 3: On page 3, in the first paragraph on lines 99 and 100, the authors describe how the spectra were acquired (in steps of 10 nm from 400 nm to 700 nm). However, they do not mention the model, company or spatial resolution of the instrument used.
Response 3: The Nikon D5100's sensitivity spectra and the Munsell color chips' reflectance spectra are publicly available. The sensitivity and reflectance spectra of the available datasets were sampled in steps of 5 nm and 1 nm, respectively. The research works on spectral recovery in the literature use a wavelength step of 10 nm, so the data of the sampling wavelengths are obtained from the datasets without interpolation. Please see the Data Availability Statement (lines 497-499) for details.
Comment 4: In radiometric terms, the spectral transmittance in Equations (4) and (7) should be defined by τ(λ). Also, why did the authors define the spectral filter as 'super-Gaussian'?
Response 4: The journal Optics adopts “Palatino Linotype” font. If a symbol has a subscript, it is best to use an uppercase symbol unless the symbol is tall enough. Therefore, we replace the suggested symbol τ by T in the revised article.
The spectral transmittance of a bandpass filter is usually bell-shaped. The super-Gaussian function is bell-shaped and is adopted because it has only three parameters besides the maximum transmittance. As emphasized in Section 3.2, the optimization problem is highly nonlinear and a brute-force grid search method is used to find the optimal solution. Other more complicated functions can be used, but will be very difficult to optimize. The same is true for the notch filter. However, from the numerical results, the accuracy of the spectrum reconstruction mainly depends on the wavelength of the spectral sensitivity lobe with reasonable bandwidth.
Comment 5: The authors of Figure 6 present histograms showing that the metrics values are better for the RGBF camera approach in a large percentage of test samples. They may be able to comment on this.
Response 5: In the revised article, Figure 6 becomes Figure 7. Note that the values of ERef, ΔE00, and SCI are the smaller the better. The value of GFC is the larger the better. From Table 2, it can be seen that the mean metrics are good for the four cameras considered in this figure. The histograms show that the metrics for the RGBF camera are good, but not better than using ISGF. More comments on this figure are added in lines 407-413.
Comment 6: In Fig. 12 and Fig. 13, what are the differences between Spectral Power Density and Spectral reflectance? How is the first one determined?
Response 6: In the revised article, Fig. 12 and Fig. 13 become Fig. 13 and Fig. 14, respectively. The Spectral Power Density is the radiant power reflected from the color chips before reaching the camera, which is the reconstructed reflection spectrum SRec using II-LUT method. The Spectral reflectance vector SRefRec is the vector SRec divided by the source spectrum vector SD65 element by element, which is described in lines 124-126.
Comment 7: In Figure 1, Maybe the figure caption should be changed by “Color imaging by using two different techniques…” in instead of “Schematic diagrams showing…”
Response 7: The figure caption has been revised according to your comment to make it better.
Comment 8: In LED, linewidth of the output spectrum, Δλ, is defined as width between half-intensity points. Maybe, terms ΔλH can be defined in similar way.
Response 8: We have replaced the halfwidth ΔλH and half separation ΔλHS by the full width ΔλF and full separation ΔλFS. Their values shown in Table 1, Table 3, and Figure 12(a) are also updated.
Additional Comments:
This phrasing is confusing: 'The same as Fig. 4, except ...'
Response: This phrasing is the caption of Fig. 5 in the original article (Fig. 6 in the revised article). Such a phrasing is intended to simplify a figure caption and avoid redundancy in wording in published papers. It can also help readers quickly identify the difference between similar conditions of two figures.
Reviewer 3 Report
Comments and Suggestions for AuthorsNice manuscript. We do not see any errors, mistakes or flaws. The rare case of the manuscript which can be accepted immediately just as it is.
Author Response
Comment:
Nice manuscript. We do not see any errors, mistakes or flaws. The rare case of the manuscript which can be accepted immediately just as it is.
Response:
Thank you for taking the time to review our article and for giving us a high rating. We hope that our article will be useful to the scientific community.
Reviewer 4 Report
Comments and Suggestions for Authorsthe comments for improvement can be found directly in the text of the manuscript sent back to you (see appendix) .
a) figure 6: on the y-axis you brought the term counts: pls. explain it, is the counts the number of the samples within 1066 Munsell colors ?
b) Figure 12: on the y-axis you brought the parameter "spectral power density": it must be clarified what is this ? in this context it is the radiant power reflected from the color sample before reaching the camera
Comments for author File: Comments.pdf
Author Response
Thank you for taking the time to review our article and for giving us a high rating. We hope that our article will be useful to the scientific community.
Comment 1) Figure 6: on the y-axis you brought the term counts: pls. explain it, is the counts the number of the samples within 1066 Munsell colors ?
Response 1: Yes, it is the number of the samples within 1066 Munsell colors. Based on your comment, we added a description of the y-axis in line 284-285. In the revised article, Figure 6 becomes Figure 7.
Comment 2) Figure 12: on the y-axis you brought the parameter "spectral power density": it must be clarified what is this ? in this context it is the radiant power reflected from the color sample before reaching the camera.
Response 2: Yes, it is the radiant power reflected from the color sample before reaching the camera. Based on your comment, we added a description of the y-axis in line 426-427. In the revised article, Figure 12 becomes Figure 13.
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript has been significantly revised and improved, incorporating the recommendations suggested at review.
Author Response
Comment:
The manuscript has been significantly revised and improved, incorporating the recommendations suggested at review.
Response:
Thank you for taking the time to review our article and for giving us a high rating. We hope that our article will be useful to the scientific community.
