Efficient Reconstruction of Low Photon Count Images from a High Speed Camera
Round 1
Reviewer 1 Report
The paper has the potential to appeal to a broad audience.
A few suggestions:
CMOS SPAD image sensors can be run in a variety of modes. Some are backside illuminated (BSI). Some have fill factor limitations. Briefly describe the sensor and explain the mode that was used. I think that you just used it as a photon counter. Did you use or disable any of the features on this particular sensor? If the sensor was not BSI, did it have have microlenses?
Is the caption for Figure 5 correct (5 instead of 20 perhaps)?
Your explanation of Bayesian retrodiction could be improved for the non-image processing audience particularly in regards to Figure 3 and equation 1- it is features so prominently in your approach.
Author Response
See attached file.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper experimentally verifies the use of Bayesian retrodictions for fast imaging with such sparse photon data. Various other image post-processing techniques are also explored and some of which result in a similar quality of image reconstruction to Bayesian retrodiction, with lower computational load.
This is a well-written paper containing interesting results which merit publication. For the benefit of the reader, however, a number of points needs clarifying and certain statements require further justification. There are given below.
1. Authors should give clear suggestions in a table or other appropriate form about the application of different methods in sparse photon imaging, including imaging quality, speed, computational cost etc., which will help readers quickly choose the method they are interested in, or make trade-offs in practical applications.
2. In Fig 8 and Fig 9, authors presented the quantitative comparison of different reconstruction methods using sparse photon data from the SPAD camera, however, visual comparison results, such as denoising or reconstructed images by different methods, are lacking.
3. In Fig 1 or Section2, authors should give an intuitive experimental setup rather than just test image.
4. From Figure 4, it is difficult to see that Bayesian retrodiction is advantageous, authors should use more intuitive and explicit experimental results to show the superiority of Bayesian retrodiction, although its indicators in SSIM and MSE are better than the original image.
Author Response
See attached file.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments to the Author
The manuscript points out that at 5 high frame rates and low incident photon flux, image post processing can provide better grayscale 6 information and spatial fidelity of reconstructed images than simple frame averaging.
As a review, the organization and hierarchy are clear and distinct. The full manuscript is very comprehensive and easy to understand. In my opinion, the paper is largely written well, and there are some specific comments which might help authors to improve the contents of this manuscript.
Comment 1. In the abstract, the author only points out the advantages of the proposed method in general. It is better to use the specific values of evaluation indicators to demonstrate superiority of the proposed method.
Comment 2. In the introduction, the novelty of the proposed method is not mentioned. It is better to specify the innovations of the proposed method.
Comment 3. In the introduction, the author only briefly listed the relevant works. It is recommended to explain and compare with each other, point out the shortcomings of the existing methods, and then lead method proposed in this manuscript. Please evaluate the performance level of the experimental results of this method compared with other studies.
Comment 4. In the introduction, it is recommended to explain its innovation and contribution after mentioning the proposed method.
Author Response
See attached file.
Author Response File: Author Response.pdf
Reviewer 4 Report
Johnstone et al. evaluate the performances of different imaging reconstruction algorithm when the photons were limited. The manuscript, however, needs a few more details to deliver a clearer picture to the readers.
Major comments:
1. The authors claim the computational cost is the major issue for the application of Bayesian retrodiction, it is very necessary for the authors to quantify this cost rather than simply say it is challenging.
2. It is very straightforward to predict that longer exposure time will make the reconstruction easier, why did authors pick 0.5 us and 2.0 us as the exposure times? Are those some critical values?
3. What is the brand of SPAD camera used in this experiment? Is 0.5 us the lowest exposure time that the image can be reconstructed?
Minor comments:
1. Line 178 “to shot to shot” to “to shot”.
Author Response
See attached file.
Author Response File: Author Response.pdf
