Image Super-Resolution Based on CNN Using Multilabel Gene Expression Programming

Round 1

Reviewer 1 Report

Comments­­_ Image Super-resolution Based on CNN Using Multi-label Gene Expression Programming

 

 

The authors have integrated MGEP as an efficient classification algorithm with SRCNN to obtain high quality restored images. However, the manuscript needs to address the following points before it can be published.

 

Major comments:

The authors should compare the MGEP-SRCNN parameters (PSNR/SSIM) among the datasets with different scales (2x,3x,4x) for a robust analysis (see Table 1). The authors should also enlarge their data set to include images of Urban 100 and Sun-Hays 80 for comparison. The authors should quantitatively discuss the increased efficiency of MGEP for the different data sets shown in Table 1 (for eg. How long does it take for super-resolving 2x,3x,4x images of the different data sets with specified computer metrics such as processing speed, RAM

 

 

Minor comments:

 

Abbreviations such as PSNR and SSIM should be expanded Authors should include the scale factor in Table 1 for each data set.

 

Author Response

Point 1: The authors should compare the MGEP-SRCNN parameters (PSNR/SSIM) among the datasets with different scales (2x,3x,4x) for a robust analysis (see Table 1). The authors should also enlarge their data set to include images of Urban 100 and Sun-Hays 80 for comparison. 


Response 1: We enlarge our dataset to include images of Urban100 and add the corresponding PSNR/SSIM results in Table 1, 2, 3. In addition, we add Figure 8 to give SR results of Urban100. We make a robust analysis with different scales (2x, 3x, 4x) in Table 1, 2, 3.

Point 2: The authors should quantitatively discuss the increased efficiency of MGEP for the different data sets shown in Table 1 (for eg. How long does it take for super-resolving 2x,3x,4x images of the different data sets with specified computer metrics such as processing speed, RAM.

Response 2: We add Table 4 to quantitatively discuss the increased efficiency of MGEP for the different data sets and give detailed discussion in Line 298-299.

Point 3: Abbreviations such as PSNR and SSIM should be expanded Authors should include the scale factor in Table 1 for each data set.

Response 3: We expand abbreviations of PSNR and SSIM in Line 90 and 254. We include the scale factor “2x, 3x, 4x” correspondingly in the title of Table 1, 2, 3.

Reviewer 2 Report

In this work, Tang et al present an improved methodology to reconstruct Super Resolution images by combining SRCNN and MGEP. The new workflow reduces the complexity of neural network parameters by replacing K means with MGEP classification thereby improving throughput in image reconstruction and resolution improvement. Tang et al., experimentally verify the improvements to image restoration by the MGEP algorithm. I felt that overall the paper was well written, with sufficient supporting information. While it may not be required, I felt that line-histograms on images in Fig.5-7 may help the reader to better appreciate the superiority of the MGEP-SRCNN algorithm (in addition to table and images).

I noticed a few minor corrections:

Pg1. Line 13 “have achieved good “Image” restoration effect. Image is missing.

Pg.1 Line 42 Image is missing in super-resolution reconstruction.

Author Response

Point 1: While it may not be required, I felt that line-histograms on images in Fig.5-7 may help the reader to better appreciate the superiority of the MGEP-SRCNN algorithm (in addition to table and images).

Response 1: We add Figure 8 to give SR results of new dataset Urban100. In addition, we add Figure 9 as line-histograms to compare the image restoration results of six algorithms and give detailed discussion in Line 337-340.

Point 2: Pg1. Line 13 “have achieved good “Image” restoration effect. Image is missing. Pg.1 Line 42 Image is missing in super-resolution reconstruction.

Response 2: We correct the mistakes and thank you for your careful review.

Round 2

Reviewer 1 Report

The authors have addressed all the concerns raised earlier. Therefore the manuscript can be accepted in its present form.