Learning-Based Text Image Quality Assessment with Texture Feature and Embedding Robustness

Round 1

Reviewer 1 Report

This paper proposes an unsupervised text image quality assessment framework. Some comments are listed as follows:

1. The effectiveness of TIQA is very related to the adopted STR model. Does the proposed TIQA method can be generalized to the other STR models?
2. In Section 2.2, the two kinds of quality inputs should be explicit.
3. Many details are missing, such as the parameters settings in Eq. (5), etc.
4. To give readers a better understanding of this field, DNN based IQA methods for various image multi-modalities are recommended to be reviewed, including Deep multi-scale features learning for distorted image quality assessment (2D), Dual-stream interactive networks for no-reference stereoscopic image quality assessment (3D), Blind omnidirectional image quality assessment with viewport oriented graph convolutional networks (VR), Interpreting representation quality of DNNs for 3D point cloud processing (PC), Blind quality assessment for image superresolution using deep two-stream convolutional networks (SR), etc.
5. It is suggested to provide some ablation tests to further verify the performance of the proposed framework.

 

Author Response

Dear Reviewer,

 

Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in Word file and my revisions in the re-submitted files.

 

Thanks again!

Author Response File: Author Response.pdf

Reviewer 2 Report

Reading your work I get the impression that you have achieved a new quality using familiar methods. I am trying to find out, to understand what is the actual contribution to the described in your achievements.

The presentation and analysis, of the methods used seems cursory. For example, you describe, the use of the OTSU method, which could be successfully substituted alternatively methods according to publications :Michalak H., Okarma K.: Robust combined binarization method of non-uniformly illuminated document images for alphanumerical character recognition. Sensors, vol. 20 no. 10, article no. 2914 (Special Issue "Document-Image Related Visual Sensors and Machine Learning Techniques"), 2020 and Michalak H., Okarma K.: Improvement of image binarization methods using image preprocessing with local entropy filtering for alphanumerical character recognition purposes. Entropy, vol. 21 no. 6, article no. 562 (Special Issue "Entropy in Image Analysis II"), 2019.

Everything seems to be correct, the research, the descriptions of the research - but there is still a sense of insufficiency, there is no clear indication of what you have achieved, the mention in the conclusion is too laconic. I would expect more development of the topic in individual chapters.

Author Response

Dear Reviewer,

 

Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in Word file and my revisions in the re-submitted files.

 

Thanks again!

Author Response File: Author Response.pdf

Reviewer 3 Report

Abstract:” real scene” could be “real-scene”

Very short sentences in your abstract!

When spelling out words, be consistent, either capitalize each word  or do not, look at your abstract.

Again, just gong over your abstract, although the topic seems interesting, it is poorly written, with very short and incomplete sentences.

The findings of your study is not clear while going over abstract and also what you did compared with previous studies. Do not use firstly, …thirdly…., this is not a story. Make it more coherent while staying away from this type of form.

Line 29, why “Text Image Quality Assessment” each word is capitalized?

You have introduction which includes literature like “A large number of subjective and objective image quality assessment (IQA) studies 36 have been proposed during recent years [13].” And also you have related works, why it is like that? Make your paper coherent and structured.

Contributions is better to go in a new subcategory,

Pseudocodes and GitHub codes should be included!

I have mainly concerns regarding the structure of your paper. Please proof edit and restructure your paper. The above are just few general comments, and you need to make sure you have gone over the whole paper to make sure all is good. Spend time to prevent future rejection. Go over my comments and answer one by one.

Regards,

Author Response

Dear Reviewer,

 

Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in Word file and my revisions in the re-submitted files.

 

Thanks again!

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed my comments.

Reviewer 3 Report

Addressed!