Semi-Coupled Convolutional Sparse Learning for Image Super-Resolution

Round 1

Reviewer 1 Report

The authors presented as interesting semi-coupled convolutional sparse learning (SCCSL) method for image super resolution. In order to formalize their problem, they put in question the assumption that coefficients in the same position from low and high images are considered equivalent which enforces an identical structure between low- and high-resolution images. Experiments results showed very promising results.

The paper is really interesting and the paper is easy to follow, however I have some minor concerns to be improved:

 

Some notations are not explained to the neophyte such as Frobenius, L1 norm etc, that should be included in the text-

Some typos are present in the text and the English should be further checked 

Is it possible to apply supervised dictionary learning techniques for image resolution? what about deep models? have you tried to compare to any of them ?

Supervised dictionary learning techniques should be pointed out in the introduction, such techniques are used for classification, please include the following papers:

 A comprehensive overview of feature representation for biometric recognition (Multimedia Tools and Applications 2018, p. 1-24)

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents a modification of the convolutional sparse coding method for single image super-resolution task. Algorithms are described in detail and the testing results are presented. The results show a small but noticeable increase in quality using the PRNS and SSIM criteria with minor increase in running time over previously reported algorithms.

Minor errors:

Line 161. “178×178 overlapped subimages”. Should be “178×178 pixel overlapped subimages” Line 177. “But when KL < 15” replace with “But when KL > 15” Line 218. “laboratory for image and video engineering”. It is unclear which laboratory is that.

The description of the experiment is unclear. Set69 is mentioned with reference to Dong et al [49]. But [49] doesn’t mention Set69, Dong et al use a training set of 91 images, and Set5 and Set14 for evaluation. Authors should clarify what their Set69 contains, the number and the size of images, and how many training subimages were derived from it.

The authors describe time complexity of the testing phase. It would also be interesting to know how long the entire training takes in their experiment.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.