Improved Low-Light Image Feature Matching Algorithm Based on the SuperGlue Net Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

• Abstract and Objectives: The abstract provides a clear overview of the study. However, it would benefit from explicitly stating the practical implications of the proposed methodology in real-world applications, such as autonomous navigation or image processing in extreme environments.

• Introduction: The introduction effectively contextualizes the problem but could further highlight the novelty of integrating MSRCR with SuperGlue compared to other state-of-the-art methods like LoFTR and Matchformer.

• Methodology: The proposed methodology is well-detailed. However, a more comprehensive explanation of parameter selection for the MSRCR algorithm and its impact on results would strengthen the robustness of the study.

• Figures and Visuals: The illustrations, such as Figure 1 and Figure 7, are informative but could benefit from higher resolution and clearer legends. Including more comparative visuals of feature matching outcomes across all experimental setups would enhance understanding.

• Results Presentation: The results are compelling, especially the quantitative improvements. However, adding a statistical significance analysis for the performance metrics would provide more credibility to the reported improvements.

• Literature Integration: The study provides a detailed review of prior algorithms but could further critically analyze how the proposed method addresses specific limitations of recent models like Matchformer or LNIFT.

• Discussion Depth: While the discussion outlines the method's advantages, it could delve deeper into the trade-offs, such as computational complexity or potential limitations in highly noisy or dynamic lighting conditions.

• Conclusion Clarity: The conclusion succinctly summarizes the findings but could better emphasize the future directions and scalability of integrating MSRCR and SuperGlue into larger systems like SLAM or robotics.

• Real-World Relevance: The manuscript focuses on low-light image matching, a critical challenge. A brief discussion on how the methodology could be adapted for other adverse conditions like fog, glare, or motion blur would broaden its impact.

• References and Citation: The references are extensive and relevant. However, including more recent works on Transformer-based approaches for image matching might reflect the latest advancements in the field.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Contribution should be clear and precise.

How grayscale variations around image feature can be improved? Mention specific method which is being used in this study.

What tool and techniques is used in this study for the image enhancement is not clear.

What is the real life application this study that is not clear, because any in any CV study we have to consider the high resolution image , then why do we need  this study.

Figure 13 is not clear and some other figures also should improved.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

Reviewer Comments

The current feature extraction operators struggle to extract high-quality features from low-light images. To address this, the authors combined multi-scale Retinex with color restoration with SuperGlue. However, this work appears to simply combine the two methods without showcasing any methodological innovation or improvement.

1. The related work section lacks a comprehensive review of research progress in the field of low-light image enhancement.

2. The datasets section does not provide sufficient details regarding the configuration of image exposure parameters. It is necessary to clarify whether experiments were conducted with varying exposure parameters.

3. The experiments lack a thorough comparison with state-of-the-art (SOTA) methods, making it difficult to validate the effectiveness of the proposed approach.

4. The quality of experimental result images is low, with blurry comparisons that fail to clearly demonstrate the outcomes. Figures 8 and 9 have overlapping numerical annotations, which negatively impact readability.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

This article presents an improved feature matching algorithm for low-light images by integrating the Multi-Scale Retinex with Color Restoration (MSRCR) with the SuperGlue network model.

Strengths: The paper introduces an innovative method that combines MSRCR with SuperGlue, aimed at solving the problem of feature extraction and matching under low-light conditions, showing high innovativeness. Through detailed experimental data and comparative analysis, the effectiveness of the proposed method is verified, showing significant improvements in the number of feature points and matching accuracy. Moreover, the article elaborates on the Retinex theory and the workings of the SuperGlue network, explaining how MSRCR technology is used to enhance image features for further matching. The potential applications in autonomous driving and robotics technologies, which have strong practical application value, are also discussed. Additionally, the diagrams and images in the article are clear and contribute to an easier understanding of the methods and experimental results, enhancing the paper's expressiveness and intelligibility.

 

Weaknesses: Although the paper introduces an innovative approach for feature extraction and matching under low-light conditions, it still has some shortcomings. The article mentions difficulties in dealing with poor quality and blurry low-light images but does not explore in depth how to improve these challenges further. Additionally, MSRCR and SuperGlue still function as two separate modules rather than a cohesive enhancement matching framework, which somewhat limits the method's overall effectiveness. Future research could focus on more profound integration and optimization to enhance system performance and broader applicability.

Issues with the Paper:

The introduction of the paper does not provide a comprehensive overview of the historical background and related work on low-light image feature matching. It should more thoroughly discuss the development of this field and existing achievements to help readers better understand the significance and contributions of this study. Moreover, the article lacks specific examples of practical scenarios, which could demonstrate the practical applications and importance of low-light image feature matching, thereby enhancing the paper's practical value and persuasive power. Adding this content would improve the overall quality of the paper, making it more readable and impactful.

The description of the steps is not detailed enough: Steps (1) to (4) are described too briefly, especially regarding the specific implementation of these steps. It is necessary to provide a detailed derivation of formulas and explain the physical significance of each formula in words. Each step's specific implementation method, including used parameters and computation processes, should also be detailed.

There is a need for a detailed description of the selection criteria and preprocessing methods for the dataset. For instance, why the ScanNet dataset was chosen and how the exposure was adjusted should be clarified. A deeper analysis of comparative experimental results is required, explaining the reasons for the performance differences among different experimental groups. A full explanation of why the combination of MSRCR and SuperGlue provides better results is needed.

The discussion section of the article is too brief and does not delve deeply into the strengths and weaknesses of the proposed method, nor does it address potential issues and directions for improvement in practical applications. It is recommended to expand this section to explore the method's effectiveness, limitations, and adaptability to specific application scenarios more comprehensively. Additionally, any potential technical challenges and future research directions should be highlighted to help advance further studies and methodological optimizations in this field.

The conclusion of the paper is too succinct and fails to adequately summarize the main contributions and findings of the research, nor does it elaborate on their importance in the related field. The conclusion should more thoroughly outline the key outcomes of the paper and emphasize their significance in the field. It should also briefly describe the potential impact and value of the research method in practical applications, providing readers with a better understanding of the research's practical prospects and actual benefits.

Furthermore, some diagrams lack clear annotations and explanations, which could impede reader understanding. More detailed annotations and explanations should be added to the diagrams to aid comprehension.

Author Response

Please see the attachment.

Author Response File: Author Response.docx