High-Capacity Data Hiding for ABTC-EQ Based Compressed Image
Round 1
Reviewer 1 Report
Thank you for submitting your original paper “High-Capacity of Data Hiding Based on ABTC-EQ”. In this case, there are many questions from this paper. Therefore, I cannot judge considering all contents. In this case, the only Section 1 and Abstract have finished reviewing. You should submit again after revising according to comment.
Abstract
- For paper title, you should revise including “Image”.
- You say, “ABTC-EQ is represented a form of trio(Q1, Q2, [Q3], BM) where Q is quantization levels (Q1 ≤ Q2 ≤Q3), and BM is a bitmap).” However, it is not enough to represent because there are many abbreviations.
- You say, “As an experiment result, our proposed scheme proves that the performance is good in the aspect of embedding capacity and quality of an image.” However, this representation is lack for persuasiveness because there is no numerical value.
- Introduction
- Because there is “Related Works” in this paper, you do not need to describe in detail, or you should merge with “Related Works”.
- The only explanation of words, sentences, and abbreviations are not enough to understand for most of readers. You should show using Figure what these words, sentences, and abbreviations are meaning.
- You say, “In this paper, we propose a new DH method based on ABTC-EQ with high embedding rates and good image quality. As we know that, because the existing DH based on AMBTC is that the quality of cover image is not high compared to original image, there is a problem the PSNR of the cover image decreases below 30dB when we embed many bits into the cover image, and we replace AMBTC with ABTC-EQ in order to solve this problem. Our purpose is to increase the performance of DH by applying LSB method to the pixels of quantization levels. There exists a demerit to generate some noises in encoding procedure, though LSB substitution method is enough to achieve our purpose. In order to solve to the problem, we propose a way to optimize the errors happening in the way and enhance the quality of an image by using Optimal Pixel Adjustment Process (OPAP) [6] in this paper.” Why you set as there is a problem the PSNR of the cover image decreases below 30dB when we embed many bits into the cover image?
- You say, “The rest of this paper is organized as follows. Section 2 briefly reviews AMBTC, ABTC-EQ, LSB subtitution and OPAP, and Ou & Sun’s method. The proposed new DH based on ABTC-EQ is introduced in Section 3. Experimental results and discussions are given in Section 4. And Section 5 sketch some conclusions.” A part of these representations is including miss types, for an example, “subtitution”. This is “substitution”.
Author Response
1. For paper title, you should revise including “Image”.
=> According to reviewer’s the comment, we modified the title as “High-Capacity Data Hiding for ABTC-EQ Based Compressed Image”.
2. You say, “ABTC-EQ is represented a form of trio(Q1, Q2, [Q3], BM) where Q is quantization levels (Q1 ≤ Q2 ≤Q3), and BM is a bitmap).” However, it is not enough to represent because there are many abbreviations.
=> Based on the reviewer's comment, we unified into one in respect of many duplicated abbreviations in Sections 1, 2.1, 2.2, and 2.3. The following are some of them.
=> In Section 2.1, we use a trio , where , to represent an AMBTC block. A trio is enough to represent AMBTC. On the other hand, in ABTC-EQ (described in Section 2.2), we first subdivide blocks to two types (non-edge blocks and edge blocks). For non-edge blocks, we still use AMBTC block (note: we here adopt the notation to distinguish the bitmap notation used in edge blocks). For edge blocks, we classify a block to three clusters, such that there are three average pixel values in these three clusters denoted as . Thus, in edge blocks, we use to represent an edge block. Finally, ABTC-EQ may be represented as or . Therefore, we use both types of blocks, one is and the other is . And meanwhile, the value of f (0 or 1) represents edge blocks or non-edge blocks.
All the above descriptions are already briefly described in Section 2.1 and Section 2.2.
3. You say, “As an experiment result, our proposed scheme proves that the performance is good in the aspect of embedding capacity and quality of an image.” However, this representation is lack for persuasiveness because there is no numerical value.
=> There is no intuitive formula or number for using reviewer's suggestions in this paper. In Section 4, we showed obviously that the performance of our proposed method was excellent through the enough
4. Because there is “Related Works” in this paper, you do not need to describe in detail, or you should merge with “Related Works”.
=> In “Introduction” section, the basic explanation of data hiding was summarized, various data hiding methods based on AMBTC were briefly reviewed, and “Related Works” section included the explanation of the necessary knowledge before entering the main subject. Therefore, the “Related Works” section is a necessary section. We need a reviewer's understanding of the validity of the reasons we've described so far.
5. The only explanation of words, sentences, and abbreviations are not enough to understand for most of readers. You should show using Figure what these words, sentences, and abbreviations are meaning.
=> We explained the details about our proposed method in Figure 3. We think the abbreviations are sufficiently understandable through Figure 3.
6. Why you set as there is a problem the PSNR of the cover image decreases below 30dB when we embed many bits into the cover image?
=> As described in Sections 2.1 and 2.2, the existing grayscale image is compressed. For example, one block (4x4) of a grayscale image consists of 16 pixels. Each pixel is composed of 8 bits, and the lower 4 bits can be used for data concealment. In the case of (512x512) image, even if concealing 1048576 bits, it can maintain more than 30dB. On the other hand, if one block of the image compressed by AMBTC is represented by a trio (Q1, Q2, BM), it is difficult to conceal a sufficient amount of data because it consists of two 8-bit pixels and one bitmap composed of 16 bits. As shown in Table 3, AMBTC-based Lena image (file size: 65KByte) represents PSNR = 33.42dB. Since the quality of the already compressed image is 33.42 dB compared to the original image, finding a way to conceal more data is substantially challenging.
7. You say, “The rest of this paper is organized as follows. Section 2 briefly reviews AMBTC, ABTC-EQ, LSB subtitution and OPAP, and Ou & Sun’s method. The proposed new DH based on ABTC-EQ is introduced in Section 3. Experimental results and discussions are given in Section 4. And Section 5 sketch some conclusions.” A part of these representations is including miss types, for an example, “subtitution”. This is “substitution”.
=> Thank you for your comment. We have correctly modified the words you mentioned. And we have also corrected the missing words.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
It's been a long time since an LSB data hiding/watermarking scheme has come to me for review. The work is well presented, with an adequate review of the corresponding literature. Presentation is clear and results are as expected. However there are other, older LSB methods with superior results to those presented. The choice of works (17-19) are not of the best choices in the field, despite being recent ('15-'18). But the most important problem for me is the choice of the embedding space. The subject has been a subject of intense work from the very early years of watermarking/data hiding. As a result there is nothing really interesting left for someone to tell for LSB embedding, classification of blocks to smooth or not with edge detectors etc.. Minor improvements over 2-3 methods which do not belong to the top self, do not provide enough justification for a good journal publication.
There are also some minor problems in the text. For example in line 82 the symbol m is used for the block size while in eq.1, k is used instead.
In lines 82-83 what do the authors mean with "without non-overlapped"
In equation 3, the inequality of the second sum I believe should be reversed
In figure 5, the horizontal axis is "bpp" but the numbers are clearly not bits per pixel. From what I've read earlier, the whole analysis is about bits, speaking of thousands secret bits. So what is really the quantity measured in the horizontal axis?
Author Response
1. There are also some minor problems in the text. For example, in line 82 the symbol m is used for the block size while in eq.1, k is used instead.
=> Thank you for your advice. According to your comment, we modified k in Equation (1) to m.
2. In lines 82-83 what do the authors mean with "without non-overlapped".
=> Thank you for your advice. The “without non-overlapped” is an improper Thus, we modified the sentences as follows:
=> "In AMBTC, the image is first divided into non-overlapped blocks sized m´m, where m´m may be set to (4x4), (8x8), (16x16), and so on."
3. In equation 3, the inequality of the second sum I believe should be reversed.
=> Thank you for your advice. According to your good comment, we corrected it as: Eq.(3)
4. In figure 5, the horizontal axis is "bpp" but the numbers are clearly not bits per pixel. From what I've read earlier, the whole analysis is about bits, speaking of thousands secret bits. So what is really the quantity measured in the horizontal axis?
=> The number of bits used for data hiding is from 20000 to 340000. We modified the caption of the x-axis from "bpp" to "bits".
Author Response File: Author Response.pdf
