Skip to content

Latest commit

 

History

History
132 lines (109 loc) · 6.76 KB

NEWS.md

File metadata and controls

132 lines (109 loc) · 6.76 KB

Imagewmark NEXT:

Breaking:

  • The default key for Watermark encoding has changed, it was previously unset

Imagewmark 0.5.0:

Added:

  • Added the --cornersync=<on|off|auto> option to control the corner sync algorithm during extraction
  • Added developer documentation, including a description of the architecture
  • Added a manual page to the documentation (imagewmark.1)
  • Added a new tests/README.md to provide an overview and usage instructions for test runs
  • Added information about test image datasets to the test documentation
  • Added an Alpine CI workflow for automated builds and checks using GitHub Actions
  • Added an IRC bot for CI build status notifications
  • Added unit tests to verify basic functionality before/after installation
  • Added new script (gen-tests-mk) that creates out-of-tree test suites of custom sizes
  • Added support to utilize all CPU cores when running test batteries
  • Added specific tests for high-resolution images and downscaling attacks

Breaking:

  • We now read all images with OpenCV instead of PIL, this potentially changes the set of supported images
  • We assume OpenImageIO/imageio.h and gcrypt.h header files are present
  • Dependencies: libgcrypt20-dev libopenimageio-dev libopencv-dev
  • Watermark encoding has changed in the 0.5.0 version due to PRNG enum modifications
  • We now build PDF versions of the docs only if pdflatex is installed

Changed:

  • Improved test realism by randomizing attack resolutions and shuffled input image selection
  • Enhanced test reproducibility with improved and deterministic seeding mechanisms
  • Moved all C++ source code into a dedicated cxx/ directory
  • Better error handling in build scripts, with improved logging for debugging
  • Major overhaul of the build system with unified Makefiles and improved dependency handling
  • Fixed various bugs in the build system and C++ code
  • Introduced installation check and uninstallation rules (make install, installcheck, uninstall, distcheck)
  • Migrated version handling to a shell script with git integration
  • Fixed image reading to properly handle subimages and miplevels
  • Added check that secret key files are only readable by the user
  • We support parallel installations of different major.minor versions

Removed:

  • Removed unused OpenCV configure script
  • Removed auto config, compile sources in subdir

Contributors

Thanks to everyone who made this release happen!

  • Tim Janik (@tim-janik)
  • Stefan Westerfeld (@swesterfeld)

For full details, see the commit history

Imagewmark 0.4.0:

  • Improved documentation in various places.
  • Enhanced test reporting with additional columns, titles, and sorting.
  • Extended and separated attack vector generations in the test suite.
  • Extended visualizations when following along the detection algorithms.
  • Extended performance monitoring by adding CPU time and wall clock measurements to various phases.
  • Extended JSON support for reporting detected watermarks.
  • Introduced --trace-quality option, to provide scores during embedding.
  • Improved test reports, added list generation for failing test cases.
  • Fixed test suite crashes, improved test rules and optimized parallel processing.
  • Fixed various image processing bugs, removed PIL dependency in favor of CV2 for image IO.
  • Use float-to-int rounding conversion for watermark embedder.
  • Improve reliability for areas which contain a constant color for some colors.
    • Clip each channel to leave some headroom before embedding the watermark.
  • Introduced the imagewmark --ssim option to compute SSIM for images after embedding.
  • Improved non-blind decoding via feature based image alignment.
  • Improved non-blind decoding via ECC based sub-pixel alignment.
  • Improved image extraction performance by relying on corner sync for non-blind decoding.
  • Added support for auto-converting CMYK and RGBA images.
  • Added support for improved detection via use of the original image (non-blind).
  • Add "cornersync" algorithm, to make watermark detection more robust for some attacks:
    • Works if original/wm corners are identical (i.e. jpg, scale, jpg+scale,...).
  • Improved peaks2grid for better handling of images with extreme aspect ratios.

Imagewmark 0.3.0:

  • Change strength computation for less visible (less robust) watermarks.
  • Improved handling of greyscale images and images in CMYK (tiff format).
  • Added experimental perspective virtual grid search using --perspective.
  • Improve speed and reduce memory consumption for watermark detection.
    • Pre-scale high resolution images before detection
    • Use 32-bit floats instead of 64-bit complex values for symmetry matrix
    • Compute auto convolution using real FFT
  • Robustness improvements:
    • Dynamically adjust zoom for high resolution images
    • Use forward error correction with convolutional codes
    • Try different window sizes for local mean / local variance
  • Support fractional zoom for the watermark.
  • Improved grid search performance.
  • Researched, documented and adjusted statistical threshold for watermark detection.
  • Properly documented peaks2grid interface.
  • Added JSON output, which provides extracted watermark bits and statistical measures.
  • Reworked and added new graphical plot generations, plots can be selected via labels.
  • Added a new test suite with maximized parallelism and lots of new attack tests.
  • Added new test reports with timing statistics, error percentages, HTML and PDF generation.
  • Improved search performance without reducing robustness using JSD match threshold.
  • Improved corner case handling and stability.
  • Added watermark pattern encryption and decryption via AES key.
  • Improved memory usage by releasing auxiliary structures earlier.
  • Properly documented new dependencies and improved Python version compatibility.
  • Adjusted the Dockerfile to enable fully self contained builds and test runs.

Imagewmark 0.2.0:

  • Significantly improved robustness and detection speed.
  • Reduced memory consumptions for extraction by an order of magnitude.
  • Added Dockerfile for reproducible builds and runs.
  • Reduce size of the randomized pixel pattern for more watermarks in small images.
  • Use smooth upsampling for embedding to reduce artifacts.
  • Improve reliability for areas of white or black pixels
    • Clip luminance to leave some headroom before embedding the watermark
  • Fixed cropping of images with odd numbered dimensions.
  • Improve automated testing.
  • Use peak angles and regularities to detect a grid of embedded watermarks.
  • Allow dumping of intermediate 'peaks' stage.

Imagewmark 0.1.0:

  • Milestone 1 reached.
  • Proof of concept embedding and extraction works, based on: Local Geometric Distortions Resilient Watermarking Scheme Based on Symmetry Zehua Ma, Weiming Zhang, Han Fang, Xiaoyi Dong, Linfeng Geng, and Nenghai Yu https://arxiv.org/abs/2007.10240