Last update: March 4, 2004
NOTE: Deadline for submitting proposals for the attack schemes is 03/03/2004 by 5PM. You first need to discuss with me in person.
Attack Schemes: Note that this is OPTIONAL. Prior to submitting you attack code, you must ensure that your own embedding scheme survives whatever attack you are proposing, and discuss with me in person before e-mailing the attack code.
The following groups have submitted their schemes: GROUP 4, GROUP 7, GROUP 3, GROUP 6, GROUP 5, GROUP 9.
If you are having problems with the above TIF images, you may want to use the following 8bit/pixel resolution images.
1. Hide at least 1000 bits in 512x512 pixel, gray scale images such that
a) the embedded images are perceptually similar to the originals. i.e., the distortion introduced by embedding is not visible.
b) Embedded images are subject to JPEG compression at 50% quality factor. Embedding should survive this attack.
c) you should be able to recover the embedded data from the JPEG compressed images. You do not have access to the original images for decoding.
2. Demonstrate your project to the instructor on March 11th. I will be available in the ECI lab from 330PM-630PM that day.
3. Prepare a 5 minute presentation (no more than 5 slides) of your project for an in-class presentation on Tuesday, March 9th. In your presentation, you should highlight the interesting features of your data hiding implementation and present some typical results.
4. Submit a brief project report (March 12, 2004, by 5PM). The report should be organized as follows--label the individual sections as A, B, C, etc.
A & B should be no more than 3 pages.
a) Group Information: List the names of the participants.
b) a brief technical description of the algorithm you implemented. List any specific research papers you referred to for your project.
c) Any issues concerning your implementation. This section should be limited to 2 pages total.
d) Results: give two typical embedding results. Print both the original and the embedded images. If you are working on image-in-image hiding, you should also include the original "signature" as well as the recovered signature image.
____________________
If you are considering additional "attacks" on the images, you should talk to me *soon*. See the links below for the current attack schemes available. Deadline for submitting your proposals is Wednesday, 03/03/2004, 5PM.
Our attack encompasses three different changes to the image. The first is to mirror the image across the y-axis, second is mirroring the image across the x-axis. The final, is a combination of both, in other words, rotating the image 180 degrees. Matlab Code.
Modifies the QIM embedding strategy, so schemes that are dependent on QIM will have trouble recovering. The code makes all low frequency values even. Matlab code.
Attached are the necessary files for our attack. Please note that both "init.m" and "Script.m" are necessary in order to execute the attack correctly.
Description:
Using a constant seed (available within the code), a pseudo-random sequence is generated. The sequence is then used to traverse through DCT coefficients and choose the correct ones to attack (i.e. change enough as to destroy hidden information while preserving the original image quality). Note that the input image name is hard-coded within the attack code and must be changed accordingly.
In the following, the image dimensions are assumed to be N x M.
1) Elimination of 1/4 of the image data in the spacial domain in a region
constrained by an arbitrarily positioned block of size N/2 x M/2.
(a_RandomBlockAttack.m)2) Segmentation and repositioning of the image on N/2 x M/2 boundaries. i.e.
The input image is split into four quarter images and these are repositioned on
the image plane. One of these four quarters is replaced by an arbitrary set of
pixel data. (a_swap_q.m)3) High pass filtering. This is not a strong attack. It is not intended as
such. It will, however, effectivly corrupt data stored in the spacial domain.
(a_hpf.m)In addition you will need the following M-files: s_Resizeuptoeight.m, s_Getfreqsamples.m, s_Getfreqsamplesize.m, and s_RestoreSpacialSamples.m.
The attached file "smile.jpg" is used in the a_swap_q.m attack function.
This is salt and pepper noise and 4 large blocks obstructing the image. Size of these blocks is about 170x170 pixels. matlab code.
Removes the LSB of the quantized DCT coefficients. The main script, "attack.m" calls three other functions "lsb.m", "dq.m", and "q.m". You need to download all four files.
RED: These are in the "attack" mode.
Group 1 Brendan Murphy bmurphy@umail.ucsb.edu |
Group 2:
|
Group 3
|
Group 4 Stephanie Hellner, PRnCsS2628@aol.com Willis Hoang Mark Leonhardt Diana Nuccitelli Peter Noyes |
Group 5
|
Group 6: Talayeh Saderi t_saderi@umail.ucsb.edu Terrence Cole
|
Group 7: Jeff [02/12 11am] |
Amandeep Chehal aschehal@umail.ucsb.edu |
Colin Ritchie critchie@engineering.ucsb.edu Jeff Beck, jbeck@umail [02/12 5pm] |
Group 10 meet again at 245PM, 2/17 [02/12 10am]
|
|
Group 12 meet again 11am, 02/19 [02/17 1045AM] Eden Cruz, imnoterin@yahoo.com Erin Cruz Jessica McAllister Trevor Parrett Kenneth Lin
Interested in the competiion phase.
|
Group 13: meet again at 230 2/23 [02/12 430pm] Mike Hu, mhu@umail YuYuan Liu Grace Lee Chen-hao(Jeremy) Chiang Damiel Imfeld |
Looking: |
|
