A No-Reference Quality Assessment Method for Screen Content Images Based on Human Visual Perception Characteristics
Round 1
Reviewer 1 Report
The article is nicely written and easy to comprehend. The technical contents contribute to the body of knowledge. There are some suggestions that are required to be addressed:
1. In the abstract, the last sentence says that “ Experimental results on three public SCI databases show that the proposed method can achieve better performance than existing methods.” It is suggested to explicitly provide the performance improvement numbers with the proposed method.
2. The first two paragraphs of the introduction clearly explain the problem and the motivation behind the proposed work. It is explained that the existing image quality assessment (IQA) methods designed for natural images cannot be directly applied to evaluate the perceptual quality of SCIs, and therefore, there is a need for an algorithm that can evaluate the image quality of SCIs objectively and accurately. Consequently, the related work (existing practices to address the highlighted problem) is clearly explained in Section 1.1. However, the limitations of related work are described in Section 1.2. It is therefore suggested to include the limitations of related work in Section 1.1. Alternatively, a new subsection can be formed with the name “Research gap”.
3. Major steps of the proposed work in Section 1.2 are nicely articulated. However, the validation and significance of the proposed method are not present in the Introduction. Therefore, it is recommended to describe the validation mechanism. Moreover, achieved results with the proposed method along with their significance (in terms of performance improvement) should be discussed before ending the Introduction Section.
4. The proposed method is nicely described in Section 2. The presentation of all technical details are well-written and easy to understand. However, the title of section 2.1 (Multi-scale Processing Technique) is not depicted in Figure 3.
5. Section 3 presents some interesting results. The experimentation is sound and comprehensive. However, it is suggested to compare the results with recently published papers on the same topic.
6. A subsection can be added to include the limitations of the proposed method.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper tested proposed algorithms on three different databases for image quality assessment. There are some suggestions for the authors before publication.
-
The authors need to rewrite the abstract. The purpose of an abstract is to form the research background, questions, objectives, and then the methods. The authors directly describe the proposed method, which makes readers don't know why the authors discuss these problems.
-
In figure 1, add explanations and subtitles for a, b, c, and d. Do not just add one figure title.
-
Expand the introduction section. Add more background information and explain why such research questions are important, what questions the authors want to answer, and what objectives the authors want to achieve.
-
Separate “related works” from section one and write a new section for “related works.”
-
In the contribution section, the authors must use past tense for several sentences.
-
For equation one, why does the equation uses multiplication, not sum? Why alpha k is not multiplied with q k? Any specific reasons for this equation?
-
Split discussions from results.
-
Line 350. Use past tense. Also, proofread for other mistakes.
-
Write a discussion section for a better structure.
-
Is it possible to share your code on GitHub or other public repositories for other researchers to validate your work?
-
When comparing your works with other researchers’ works, it might be good to compare your method structures, not just the performance represented by scores.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript describes a new method for quality assessment of screen content images that takes into account perceptual characteristics of human vision system. Several features are proposed that are extracted from the image using dictionary learning and sparse coding, color features such hue and saturation are also utilized. The new method was compared with state-of-the-art methods using several datasets. The obtained results showed that the developed method can achieve better performance.
Though the paper is very interesting the presentation is to be enhanced before publication by major revision.
Major issues:
There are many hard to understand phrases, so the language is to be enhanced.
All abbreviations are to be decoded on first mention.
Methods are to be described with more details
I think the definition of Quality that the authors struggle to asses is to be given as early as possible in the paper.
eq.2 the dimensions of matrices and vectors are to described
line 216 what is 'the number of occurrences of atoms in the dictionary'?
eq.4 what does '*' mean here? If it is just a mutiplication than it's better to use 'x' or dot or nothing as an asterisk is more appropriate for convolution.
line 260 in line 176 it was said that D is a matrix, here it is a vector, in line 203 X capital includes x and here the notation is opposite. Please use the same notation everywhere.
Figures are to be enhanced
line 24 Fig.1 - please explicitly indicate the type of image on each panel of the Figure.
line 137 Fig.2 All symbols are to be explained such as L2, q1,..,q5, cross, downarrow.
Figure 7 the abbreviations in the legend are to be described
Results are to be presented with more details
Table 1. Are these values for training subset or testing subset, are they means or just one run? Ideally cross-validation is to be performed and mean value for performance metrics for testing samples are to be compared with statistical criteria to test the hypothesis of equal means or one-sided one that one mean is grater than another.
The computational complexity and time for different methods are to be compared and discussed.
The Conclusions are to be strengthen. The fact that some method can be better than another is rather weak.
Some specific minor issues:
line 2-5 The proposed method takes the perceptual characteristics of the human visual
into full consideration by simulating the multi-channel working mechanism of the human visual
system (HVS) through pyramid decomposition and the information extraction process of the brain
with the help of dictionary learning and sparse coding. - hard to understand, please rephrase
line 127 Then the energy feature is extracted to represent the overall quality. Secondly,
the overall quality representation is obtained by extracting the energy features of the
image. - these two sentences seem to tell the same
line 167 Dictionary learning is to capture the most essential common features among thousands
of targets, which can help restore and replace the original version to facilitate further
processing. - hard to understand please rephrase
line 172, 198 the abbreviation K-SVD, OMP is to be described and a brief outline of the algorithm
line 350 we employ a nonlinear logistic regression process is performed - please rephrase
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors sufficiently improved the text and took almost all my comments into account. Unfortunately, the comparison of methods is still performed based only on averages of metrics. As I mentioned in the previous review the results of methods are to be compared using statistical criterion such as Mann-Whitney, for example. Considering Table 1 and RMSE metric, the authors performed 1000 experiments and consequently have 1000 results for each method. The criterion will compare these two samples of 1000 elements and show the statistical significance of the hypothesis that the average for Proposed method is less then for GFM. It may happen that the difference in averages is insignificant and both methods are equally efficient.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
