sealwatch 2024.3

Implementation of modern image steganalysis algorithms

Contents

The following steganalysis feature implementations are provided:

Abbreviation	Full name	Dimensionality	Reference	Output format
SPAM	Subtractive pixel adjacency matrix	686	Reference	ordered dict
JRM	JPEG rich model	11255	Reference	ordered dict
DCTR	Discrete cosine transform residual features	8000	Reference	1D array
PHARM	Phase-aware projection rich model	12600	Reference	ordered dict
GFR	Gabor filter residual features	17000	Reference	5D array

These implementations are based on the Matlab reference implementations provided by the DDE lab at Binghamton University.

Usage

Extract features from a single JPEG image

from sealwatch.features.gfr.gfr import extract_gfr_features_from_file

features = extract_gfr_features_from_file("seal1.jpg")

Extract features for a directory of JPEG images and store them to a HDF5 file in the output directory.

python sealwatch/batch_extraction/extract_features.py \
  --input_dir input_dir \
  --output_dir output_dir \
  --feature_type "gfr"  

After having extracted features from cover and stego images, you can train an FLD ensemble as binary classifier.

import numpy as np
from sealwatch.ensemble_classifier.fld_ensemble_trainer import FldEnsembleTrainer
from sealwatch.ensemble_classifier.load_features import load_and_split_features

cover_features = "cover_features.h5"
stego_features = "stego_features.h5"
train_csv = "train.csv"
test_csv = "test.csv"

cover_features_train, stego_features_train, cover_features_test, stego_features_test = load_and_split_features(
    cover_features_filename=cover_features,
    stego_features_filename=stego_features,
    train_csv=train_csv,
    test_csv=test_csv,
)

# Training is faster when arrays are C-contiguous
cover_features_train = np.ascontiguousarray(cover_features_train)
stego_features_train = np.ascontiguousarray(stego_features_train)

# The hyper-parameter search is wrapped inside a trainer class
trainer = FldEnsembleTrainer(
    Xc=cover_features_train,
    Xs=stego_features_train,
    seed=12345,
    verbose=1,
)

# Train with hyper-parameter search
trained_ensemble, training_records = trainer.train()

# Concatenate the test features and labels
X_test = np.concatenate((cover_features_test, stego_features_test), axis=0)
y_test = np.concatenate((
    -np.ones(len(cover_features_test)),
    +np.ones(len(stego_features_test))
), axis=0)

# Calculate test accuracy
test_accuracy = trained_ensemble.score(X_test, y_test)

Feature formats

Note that the feature extractors return different formats: 1D arrays, multi-dimensional arrays, or ordered dicts. The reason is that feature descriptors are composed of multiple submodels. Retaining the structure allows the user to select a specific submodel. The following snippets show how to flatten the features to a 1D array.

Multi-dimensional array

from sealwatch.features.gfr import extract_gfr_features_from_file

# The GFR feature extraction returns a 5-dimensional array:
# - Dimension 0: Phase shifts
# - Dimension 1: Scales
# - Dimension 2: Rotations/Orientations
# - Dimension 3: Number of histograms
# - Dimension 4: Co-occurrences
features = extract_gfr_features_from_file("seal1.jpg")

# Simply flatten to a 1D array
features = features.flatten()

Ordered dict

from sealwatch.features.pharm import extract_pharm_revisited_features_from_file
from sealwatch.utils.grouping import flatten_single

# The PHARM feature extraction returns an ordered dict
features_grouped = extract_pharm_revisited_features_from_file("seal1.jpg")

# Flatten dict to a 1D array
features = flatten_single(features_grouped)

After saving a batch of flattened features to an HDF5 file, you can also re-group them.

from sealwatch.utils.grouping import group_batch
from sealwatch.utils.constants import PHARM_REVISITED
import h5py

# Load the flattened features
with h5py.File("pharm_features.h5", "r") as f:
  features_flat = f["features"][()]

# Re-group the flat features
features_grouped = group_batch(features_flat, feature_type=PHARM_REVISITED)

# features_grouped is an ordered dict. The keys are the submodel names. Each value is an array with the shape [num_samples, submodel_size].

```python
from sealwatch.utils.grouping import flatten_single

# PHARM feature extraction returns an ordered dict
features_grouped = extract_pharm_original_features_from_file(**kwargs)

# Flatten dict to a 1D ndarray
features = flatten_single(features_grouped)

# GFR feature extraction returns a 5D ndarray
features_5d = extract_gfr_features_from_file(**kwargs)

# Simply flatten the array
features = features.flatten()

Unit tests and compatibility

For notes on compatibility with previous implementation, see compatibility.md.