Next Article in Journal
Light Field Super-Resolution via Dual-Domain High-Frequency Restoration and State-Space Fusion
Next Article in Special Issue
MSP-EDA: Multivariate Time Series Forecasting Based on Multiscale Patches and External Data Augmentation
Previous Article in Journal
Reality Head-Up Display Navigation Design in Extreme Weather Conditions: Enhancing Driving Experience in Rain and Fog
Previous Article in Special Issue
Multiple Kernel Transfer Learning for Enhancing Network Intrusion Detection in Encrypted and Heterogeneous Network Environments
 
 
Article
Peer-Review Record

Color Normalization Through a Simulated Color Checker Using Generative Adversarial Networks

Electronics 2025, 14(9), 1746; https://doi.org/10.3390/electronics14091746
by Albert Siré Langa 1,*, Ramón Reig Bolaño 1, Sergi Grau Carrión 1 and Ibon Uribe Elorrieta 2,*
Reviewer 1:
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Electronics 2025, 14(9), 1746; https://doi.org/10.3390/electronics14091746
Submission received: 26 March 2025 / Revised: 14 April 2025 / Accepted: 18 April 2025 / Published: 25 April 2025
(This article belongs to the Special Issue Machine Learning in Data Analytics and Prediction)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The author proposed a color normalization method (GAN-CN-CC) based on generative adversarial network (GAN), aiming to eliminate the reliance on physical color check cards in traditional methods. However, there are some shortcomings in the manuscript:


1. The data set is too small and lacks diversity. The data set used in the manuscript in the experiment only has about 1,500 images. The training data set for the GAN network model is relatively small, and the data set is too singular to identify the performance of the model accurately.
2. In the model description part of the manuscript, the model framework diagram provided by the author in the manuscript is too general, and there is no detailed description of the proposed model. Please explain the specific details in detail.
3. The experiment lacks a large number of related experiments and comparative experiments. In the experimental part of the manuscript, the author only experimented with the proposed model GAN-CN-CC without considering other related studies and lacked relevant model control
experiments. For example, the model performance comparison experiment between traditional models, CNN-based models, GAN-based models, etc.
4. There is a lack of research on the computational cost of model training and inference. 

Author Response

Please, see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors
  • The abstract is too brief. The authors should expand the abstract and include specific results from the study to demonstrate the value of the research.

  • The manuscript lacks a review of related work. Introducing relevant studies can help readers understand the current limitations in the field and better highlight the contributions and innovations of this research.

  • The manuscript does not include any tabular data. For a study focusing on artificial intelligence algorithms, it is not sufficient to present only training curves. The authors should provide detailed performance metrics of the proposed model, such as accuracy (ACC), F1 score, number of parameters, and FLOPs. Additionally, the proposed method should be compared with both classical mainstream models and the latest relevant algorithms to validate its superiority.

Author Response

Please, see attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

They introduce a GAN-based method (GAN-CN-CC) that simulates the function of a physical color checker for color normalization in digital images. The model eliminates the need for an accurate color checker, using a Pix2Pix GAN trained on color-corrected image pairs.

The paper is interesting and valuable, but a few key points should be addressed:

  1. The GAN was trained without a validation set. Can the authors comment on how they monitored overfitting or generalization?

  2. The GAN-CN-CC model struggles with black and white tones, as noted. Could techniques like contrast-aware loss or histogram equalization help?
  3. Line 13 – “Initially, an external physical marker known as a color checker, combined with a machine learning algorithm...” → suggest changing to “Initially, a machine learning algorithm combined with an external physical color checker...” for clarity.
  4. Line 395 – “Values between 0.0 and 0.5 reflect low similarity…” → should be “Values from 0.0 to 0.5 indicate low similarity…” for smoother phrasing.
  5. How does the model perform with real-world, non-laboratory images, such as those taken in outdoor lighting or with reflective surfaces?
  6. The input image resolution is downsampled to 256×256. Would training on higher-resolution patches improve fine detail and color edge correction?
  7. Including one visual failure case and explaining why the model fails there would be helpful. This helps users understand the model's limitations.

Author Response

Please, see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have revised the questions raised.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors addressed all my suggestions.

Back to TopTop