KGBCREST

Kingry Graphical Book Cipher Compression Resistant Encrypted Steganography Tool

A steganography tool that hides secret messages in images using a book cipher combined with DCT watermarking. Messages survive JPEG compression down to quality 60.

Quick Start

Double-click run.bat - this starts the server and opens your browser to the tool.

Or manually:

python server.py

Then open http://localhost:8080

How It Works

The Book Cipher

Messages are encoded using character positions in a shared source text file (source_text.txt). Each character is found in the source text, and its relative position from the previous character is recorded.

Example: Encoding "The"

• Find 'T' at position 325 -> record: 325 (first character, absolute)
• Find 'h' at position 225 -> record: -100 (100 characters backwards)
• Find 'e' at position 228 -> record: +3 (3 characters forwards)

Result: [325, -100, +3]

The cipher always finds the nearest occurrence of each character (forward or backward), minimizing the numbers stored.

Encoding Pipeline

Message -> BOOK CIPHER -> COMPRESSION -> REED-SOLOMON -> REPETITION -> DCT WATERMARK -> Encoded Image

• Book Cipher: Characters to relative positions in source text
• Compression: zlib compresses the position data (40-60% reduction)
• Reed-Solomon: Adds 64 parity symbols for error recovery
• Repetition: Each bit repeated 7x (majority vote recovery)
• DCT Watermark: Embedded in frequency domain coefficients

DCT Watermarking

The data is embedded in the Discrete Cosine Transform (DCT) coefficients of the image - the same frequency domain that JPEG compression operates in. This is why encoded images survive JPEG compression.

By embedding data in mid-frequency coefficient (4,3), we hide information in a region that:

• Survives JPEG quantization (unlike high frequencies)
• Isn't visually obvious (unlike DC/low frequencies)
• Has enough "room" to encode bits robustly

The strength parameter (150) determines the quantization step size:

• Higher strength = more robust to compression, but more visible artifacts
• Lower strength = less visible, but more fragile
• 150 is tuned to survive JPEG quality 60 while remaining visually subtle

Capacity

Message capacity depends on image size:

Image Size	Resolution	Approximate Capacity
VGA	640x480	~75 characters
HD	1280x720	~250 characters
Full HD	1920x1080	~500 characters
4K	3840x2160	~2000 characters

JPEG Survival

Tested results:

Format	Quality	Result
PNG	lossless	Works
JPEG	90	Works
JPEG	80	Works
JPEG	70	Works
JPEG	60	Works
JPEG	55	Fails
WebP	lossless	Works
WebP	80	Works

Files

File	Description
run.bat	Start the server and open browser
server.py	Python web server backend
web_client.html	Web interface (served by server.py)
book_cipher.py	Core encoding/decoding algorithms
source_text.txt	Shared source text for book cipher

Dependencies

pip install pillow numpy scipy reedsolo

Technical Details

Settings (tuned for JPEG Q60 survival)

STRENGTH = 150      # DCT modification strength
RS_SYMBOLS = 64     # Reed-Solomon parity symbols
REPETITION = 7      # Bit repetition count

Position Encoding

• Zigzag encoding: Converts signed integers to unsigned for efficient storage
• Varint encoding: Variable-length encoding (small numbers = fewer bytes)
• zlib compression: Further reduces data size

Error Correction

1. Reed-Solomon: Adds 64 parity bytes, can correct up to 32 byte errors
2. Repetition: Each bit repeated 7 times, majority vote recovers from noise

Limitations

1. Source text required: The source_text.txt file must contain all characters in your message
2. Image modification: Cropping or resizing the image will break the watermark
3. Compression floor: JPEG quality below 60 corrupts the embedded data
4. Capacity: Limited by image size (see capacity table above)

License

Copyright (c) 2026 Adam Kingry. Licensed under the MIT License.