stegosphere 1.3

A flexible steganography and steganalysis library supporting various file types, including encryption and compression

Stegosphere

Stegosphere is a versatile library for both applying steganography (hiding data within media) and performing steganalysis (detecting hidden data). It provides a flexible framework for any data representable as a NumPy array, including images, videos, audios and TrueType fonts. Furthermore, multi-file embedding, Hamming codes, payload compression, encryption and visualization are available, as well as a research toolbox for evaluating method performance and security.

Documentation

Online documentation is available at https://maximilian-koch.github.io/stegosphere/

Detailed examples are available at https://github.com/Maximilian-Koch/stegosphere/tree/main/examples

Installation

The latest release (and required dependencies) can be installed from PyPI: pip install stegosphere

To install all optional dependencies for various additional operations, use: pip install stegosphere[complete]

The only requirements are numpy and lazy-loader. The file containers work with PIL/Pillow for images, fontTools for TTF files, cv2 for videos, and soundfile for audios. Using files containers is not required as they only provide the binding between files and numpy arrays.

Examples

Extracting a payload

The following example reads a hidden payload from the library's logo:

from stegosphere.embeddings import LSB
from stegosphere import image
from stegosphere import binary_to_data

img = image.ImageContainer('logo_small_stego.png')
px = img.read()
payload = LSB.extract(px)

print(binary_to_data(payload).decode())

Embedding and extracting

The following example hides and reads a payload from an audio file using echo hiding steganography:

from stegosphere.embeddings import echo_hiding
from stegosphere import audio

aud = audio.AudioContainer('audio.wav')
frames = aud.read()
payload = '01110100011001010111001101110100'
stego_frames = echo_hiding.embed(frames,payload,alpha=0.4)
#aud.save('stego_audio.wav', stego_frames)

dec_payload = echo_hiding.extract(stego_frames,n_bits=len(payload),alpha=0.4)
print(dec_payload)
print(payload)
#minimal differences are possible due to imperfect delay detection!

Advanced functionalities

The following example uses (Pixel) Value Differencing to hide a payload within an image. The payload uses Hamming 7-4 error correcting codes and is embedded randomly across the image with a seed. PSNR is measured and a Gaussian noise attack is being simulated. The image is saved, loaded, the bit-error-rate is measured, and differences are visualized. (Note: Storing and saving into a file is not needed for analysis. This is purely for demonstration.)

import stegosphere
from stegosphere.embeddings import VD
from stegosphere import image
from stegosphere.costs import HILL
from stegosphere.selector import greedy_indices
from stegosphere.tools import ecc
from stegosphere.analysis import imperceptibility, robustness, accuracy, visualizer

img = image.ImageContainer('logo_small.png')
px = img.read()
payload = "payload!"*300

#convert to binary so that ECC can be applied
bin_payload = stegosphere.data_to_binary(payload)
ecc_payload = ecc.Hamming7_4.encode(bin_payload)
#do payload framing / end-of-payload manually
stego_px = VD.embed(px, ecc_payload, framing=None, seed=0)

psnr = imperceptibility.psnr(px, stego_px)
#simulating attack with very low sigma
stego_px = robustness.gaussian_noise_attack(stego_px,sigma=0.1)

print(f"PSNR: {psnr:.2f}dB")
img.save('stego_image.png',stego_px)

stego_img = image.ImageContainer('stego_image.png')
stego_px = img.read()
output = VD.extract(stego_px, framing=None, seed=0)
decc_payload = ecc.Hamming7_4.decode(output[:len(ecc_payload)])
ber = accuracy.bit_error_rate(decc_payload, bin_payload)
print(f"Bit error rate: {ber:.2f}%") #should be 0%
#Due to the seed, the differences are randomly distributed.
visualizer.difference_heatmap(px,stego_px)

Contributing

Any support or input is always welcomed. Additional embedding and cost functions are very much needed.

Contact:

Email: [e-mail]

LinkedIn: https://www.linkedin.com/in/maximilian-jw-koch/