Image, Audio & Emoji Steganography

**A Python toolkit for hiding secret messages inside images, audio files, and emojis — three independent steganography techniques in one project.**

Developed as part of the AI Vicharana Shala programme at IIT Ropar (CSE Dept. × iHub-AWaDH)

Overview · Techniques · Demo · Setup · How It Works · Use Cases

---

Overview

Steganography is the art of hiding information inside other data so that its very existence is secret — unlike encryption, which hides the content, steganography hides the fact that a message exists at all. This project implements three distinct steganography techniques in Python:

Technique	Cover Medium	Method	Detection Resistance
Image	PNG / BMP	LSB (Least Significant Bit)	High — no visible change
Audio	WAV	LSB in audio samples	High — inaudible
Emoji	Unicode text	Zero-width characters	Very high — invisible

---

Demo

Image Steganography — before and after

Original image → image with hidden message embedded

Original	With hidden message

The two images are visually identical. The difference exists only at the binary level (LSB of pixel values).

Decoded output — message successfully extracted:

---

Techniques

1. Image Steganography (LSB)

Original pixel:  [10110110]  [11001010]  [10101011]
                     R           G           B
 
Secret bit:          1           0           1
 
Modified pixel:  [10110111]  [11001010]  [10101011]
                     ↑                       ↑
               bit flipped            bit flipped
               (imperceptible change of 1/255 in brightness)

• Embeds 1 bit per colour channel (RGB) per pixel
• 3 megapixel image = ~1.1 MB of hidden data capacity
• Uses OpenCV to read/write pixel values
• Extraction reverses the process — reads LSB of each channel

2. Audio Steganography (LSB)

Audio sample (16-bit):  1001101010110110
Secret bit:                             1
 
Modified sample:        1001101010110111
                                        ↑
                              last bit changed
                    (change = 1/65536 of full scale — completely inaudible)

• Embeds 1 bit per audio sample using Python's wave module
• Human hearing threshold: changes below ~0.003% amplitude are inaudible
• LSB modification = 0.0015% change — well below detection threshold
• Works on uncompressed WAV files

3. Emoji Steganography (Zero-Width Characters)

Visible text:    "Hello!"
Hidden message:  "secret"
 
Encoded:         "Hello!​‌‍​‌‍​‌‍..."
                        ↑↑↑↑↑↑↑↑
            Zero-width characters invisible to human eye
            but present in Unicode string

• Maps each character to a sequence of zero-width Unicode characters
  • \u200b (Zero Width Space) = binary 0
  • \u200c (Zero Width Non-Joiner) = binary 1
• Hidden message is completely invisible in any text display
• Useful for watermarking text documents

---

Setup

Requirements

pip install opencv-python numpy emoji

Full dependencies:

opencv-python>=4.5.0
numpy>=1.21.0
emoji>=2.0.0

Note: Audio steganography uses Python's built-in wave module — no additional install needed.

Clone and run

git clone https://github.com/imAryanSingh/Steganography.git
cd Steganography
pip install opencv-python numpy emoji
python Steganography.py

---

How It Works

Image — encode & decode

# ENCODE: hide message in image
python Steganography.py
# Select option 1 (Image Steganography)
# Enter: input image path, secret message, output image path
 
# DECODE: extract message from image
# Select option 1 → decode
# Enter: encoded image path
# Output: original secret message

Audio — encode & decode

# ENCODE: hide message in audio
# Select option 2 (Audio Steganography)
# Enter: input WAV file, secret message, output WAV file
 
# DECODE: extract from audio
# Select option 2 → decode
# Enter: encoded WAV file
# Output: original secret message

Emoji — encode & decode

# ENCODE: hide message in text using emojis/zero-width chars
# Select option 3 (Emoji Steganography)
# Enter: cover text, secret message
 
# DECODE: extract from emoji text
# Select option 3 → decode
# Enter: encoded text
# Output: original secret message

---

Project Architecture

Steganography.py
│
├── ImageSteganography class
│   ├── encode(image_path, message, output_path)
│   │     → reads pixels → embeds bits → saves new image
│   └── decode(image_path)
│         → reads pixels → extracts LSBs → reconstructs message
│
├── AudioSteganography class
│   ├── encode(audio_path, message, output_path)
│   │     → reads WAV samples → embeds bits → writes new WAV
│   └── decode(audio_path)
│         → reads WAV samples → extracts LSBs → reconstructs message
│
└── EmojiSteganography class
    ├── encode(cover_text, secret_message)
    │     → converts chars to binary → inserts zero-width chars
    └── decode(encoded_text)
          → extracts zero-width chars → converts binary → message

---

Capacity Calculator

How much data can you hide?

Cover Medium	File Size	Hidden Capacity
1MP image (PNG)	~3 MB	~375 KB of text
5MP image (PNG)	~15 MB	~1.87 MB of text
1 min WAV (CD quality)	~10 MB	~1.25 MB
5 min WAV	~50 MB	~6.25 MB

---

Use Cases

Application	Technique	Why
Digital watermarking	Image LSB	Embed copyright info invisibly
Covert communication	Any	Message existence is deniable
Document authentication	Emoji/text	Hidden integrity check
CTF / security challenges	All three	Classic steganography puzzle
Research in ML-based steganalysis	Image	Generate training data

---

Security Note

LSB steganography is a basic technique — it can be detected by statistical analysis (steganalysis tools like StegExpose). This project is educational and demonstrates the core principles. Production-grade steganography uses more sophisticated approaches (DCT domain embedding, adaptive bit allocation, deep learning-based methods).

---

Technologies

Library	Purpose
OpenCV	Image read/write, pixel manipulation
NumPy	Array operations on pixel/sample data
wave (stdlib)	WAV audio file read/write
emoji	Emoji character lookup and encoding

---

About the Author

Aryan Singh — AI/ML Engineer

Developed during the AI Vicharana Shala residential programme at IIT Ropar (May–Jul 2024)

---

Also see

• Wake-Word Detection for ISRO TRISHNA Satellite — CNN + MFCC, built at ISRO SAC Ahmedabad
• Recommendation System for Retail Stores — Collaborative filtering, IIT Ropar
• Smart Vision Quality Control — Top 0.3% Flipkart GRID 6.0