Detecting whether steganography is present in an image using machine learning.

• Least significant bit (LSB) embedding functions using various embedding location sequences.
• Generates dataset from set cover images.
• Creates feature vectors from statistical moments of autocorrelation and discrete wavelet decomposition measures.
• Exploratory plots of images and training datasets.
• Preliminary model comparisons.

Model 5-fold cross validation results on 2000 (256x256) images of cats and dogs with various message sizes embedded. Image messages are converted to strings and hidden in a cover image using the LSB algorithm.

Image type counts below. Cover is the original image and 64x64 is the image size that is hidden in a cover image.

Counter({'16x16': 350, '32x32': 333, '64x64': 317, 'cover': 1000})

• Model benchmarks for different LSB embedding generator types.
• Look at model performance for different image types (only cats and dogs at the moment).
• Model persistence for well performing trained models.
• Improve LSB embedding location sequences.
• Extend/improve features.
• Look at jpg images and embedding in discrete cosine coefficients.