SIP 2006 的 final paper

A NOVEL QUANTITY BASED ON CLIPPING STATISTICS FOR JSTEG STEGANALYSIS,
Yeuan-Kuen Lee, Shih-Yu Hwang, Zhan-He Ou,
8th IASTED Int. Con. on Signal & Image Processing (SIP 2006),
August 14-16, 2006, Honolulu, Hawaii, USA

這篇論文是根據 reviewers 的意見修改後,
再依照 SIP 研討會的論文格式編輯後的 final paper。

這次的論文修改, 主要著重在加強實驗的測試影像與描述實驗結果上。根據第二位 reviewer 的意見, 我們原本的測試影像和依般論文常用的影像不同, 所以我特地從 USC-SIPI Image Database 的 Miscellaneous 中, 下載了 8 個大小為 512*512 的標準彩色影像做測試, 並將實驗結果用 Table 1 與 2 詳細地描述呈現。

這次 reviewers 給意見, 應該可以算是對我們有所肯定, 各項分數的評分如下:

The numerical list below is a list of the reviewers' averaged score for the criteria of originality, contribution, references, presentation, and language for your paper. Please note that the final result of the paper is in no way solely dependent upon the numerical scores assigned by the reviewers. Other factors such as the total number of submissions, comments, and relevance to conference topic area also influence the final result. Also, please note that some reviewers may choose not to state any comments.

(Key: 1=Excellent, 2=Good, 3=Satisfactory, 4=Unsatisfactory, 5=Poor)

Originality: 2.0
Contribution: 3.0
References: 3.0
Presentation: 3.0
Language: 3.0

Final Result: Accepted

第一個 reviewer 的 comments 如下:

Contribution:
Describes the use of a statistical method based on color clipping to identify stego-images and to extract covert information from Jsteg stego-images.

Suggestions:
1. In the second page substitute "space than RGB space" by "space instead of RGB space".
2. In the fifth page substitute "should step 2 repeat" by "should step 2 be repeated".
3. It would be advantageous to adress more the low contrast images problem.

第二個 reviewer 的 comments 如下:

Contribution:
The paper presents a new approach in steganalysis detection for images modified using the Jsteg tool. The main contribution of the paper is that it makes the association between high number of color overflows and the presence of steganographic message. The authors then suggest that a test for hidden data can be performed by further embedding hidden data. The experimental results may or may not be illustrative. As the authrors acknowledge, the number of random message insertions may be hard to find. It is also unclear how large the message must be. The images used are not standard and have not been presented or described in detail.

Overall, I think the paper has real potential to spark a discussion at the conference. I believe however that the way the method is presented as well as the value of the experiments are working against the overall paper value. A sign of the experiment's solidity is the lack of comparison with any steganalysis tool.

Suggestions:
Clealry describe the data sets. Refine the algorithm details (how many rounds were really run for each image, what was the size of the embed message).