JPEG Steganalysis Using Machine Learning
This project’s goal is to use Binghamton’s feature extraction software to train a ML framework to detect F5
embedded payloads in JPEG images.
1. Learn about JPEG image format, JPEG steganography and steganalysis.
2. Understand F5 and how it embeds a payload into a JPEG carrier.
3. Use Binghamton’s feature extraction software to extract features from JPEG images.
4. Build a Machine Learning model that can detect the presence of F5 steganography.
You may find other goals as you progress, but the above items will allow you to design your research effort.
Specific Project Goals by Semester
Goals for Dec 2019
• MilSto01: Due in class Wed 28 Aug 2019; see syllabus for more details. Download feature extraction
software from Binghamton and F5 embedding software from GitHub. We will provide you with 10 JPEG
images and 2 documents to use as payloads.
• MilSto02: Due in class Wed 18 Sep 2019; see syllabus for more details. Convert five of the JPEG images
to BMP using ImageMagik. The original 10 JPEG images can be considered “clean”. Using F5, embed the
provided documents into the BMP images to produce 10 JPEG “dirty” images. Using the Binghamton
software extract the features from all 20 images. Instructor and mentors will help.
• MilSto03: Due in class Wed 16 Oct 2019; see syllabus for more details. Using the 20 images and the ML
framework, train your model to detect clean and dirty images.
• MilSto04: Due in class Wed 06 Nov 2019; see syllabus for more details. Demonstrate your functioning
software across a number of different covert and overt files.
• MilSto05: Due in class Wed 20 Nov 2019; see syllabus for more details. Demonstrate your refined
functional system and give an updated presentation. This presentation should include (a) your current work
refined from your MilSto04 presentation, (b) the plan you create for the Spring 2020 semester and (c) your
ideas about the additional overt images and covert payloads.
Goals for May 2020
Adding additional images with different payloads.
Continue testing and improving the effectiveness of your ML model at detecting F5 steganography over
additional overt images and covert payloads.
Presentation and demonstration at ForenSecure20 in April 2020.
projDesc-JPEGMLstego-v3.docx Page 2 of 2
As the project progresses, advanced versions of each of the deliverables will be required so that the instructor and
mentors can aid and facilitate the students’ work. This will be true for the Fall 2019 and also the Spring 2020
semesters. The four deliverables are:
Technical document (TecDocXX)
User manual (UsrManXX) for your implemented software
Demonstration video (DemVidXX) demonstrating your nicely working proof-of-principle system
Presentation (PresXX) for describing your work to the mentors, the instructor, and to an audience.
The XX is used to define your version of the deliverable. For instance, TecDoc03 would be your 3rd TecDoc
deliverable. The deliverable schedule is defined in the syllabus class schedule. You will give a presentation and
demonstration of your work at ForenSecure20 in April 2020.
Estimated Level of Effort and Challenges