No Watermark

Detect and Remove hidden watermark in the screenshot && Randomize fingerprint to avoid tracking. 100% offline.

✨ What's New

• Modern UI & Asynchronous Engine: Fully redesigned with CustomTkinter. Features True Dark/Light modes, frameless dialogs, and a non-blocking asynchronous multi-threaded backend.
• Advanced 4-Layer Detection:
  • LSB (Spatial): Scans the least significant bits.
  • FFT, DCT, DWT (Frequency): Analyzes geometric patterns in Fourier and Wavelet spectrums.
• Adaptive Clearance Modes: Aggressive (maximum removal), Balanced (recommended), Conservative (quality preservation).
• Privacy Engine:
  • Rand Name: Hashes the output file with a random 9-character string.
  • Rand Size: Invokes a micro-crop (±2px) imperceptible to the human eye to destroy resolution-based tracking.
  • Force JPEG: Compresses lossless images to strip container-level steganography.

🛠️ Usage

Quick Start (No Installation)

1. Clone the repo and install dependencies:

git clone https://github.com/AndrewWangDev/remove_watermark.git
cd remove_watermark
pip install -r requirements.txt

2. Launch the app:

python src/main.py

3. Workflow: Browse for an input image -> Select your detection/privacy parameters -> Hit Detect Watermarks (to scan) or Clean & Save (to output the neutralized image).

📊 Scoring & Verification

The Detection Engine runs a fully mathematical analysis to compute anomaly scores across 4 dimensions. Based on statistical thresholds, it determines a global probability of watermark existence:

--- Starting DETECT ---
Method:     | LSB       | FFT       | DCT       | DWT       | 
Score:      | 0.25      | 239.23    | 43.59     | 0.15      | 
Result:     | Found     | Clean     | Clean     | Clean     | 
Conclusion: The image is unlikely to be watermarked (23% probability).
            Watermark components likely present in: LSB.
--- DETECT Finished ---

• LSB (Spatial): Calculates entropy and bit-plane correlation. High randomness in LSB yields an anomaly score (e.g., > 0.2 threshold triggers "Found").
• FFT / DCT / DWT: Identifies geometric shapes or high-frequency spikes in the magnitude spectrum. A high-variance score indicates artificial manipulation.
• Global Probability: A weighted Bayesian combination of the 4 independent dimension scores, giving you a definitive percentage (e.g., 23%) of watermark likelihood.

🧠 Core Algorithm Architecture

This application employs a dual-domain analysis to expose and destroy steganographic artifacts:

    [ Original Suspicious Image ]
                  │
                  ▼
 ┌─────────────────────────────────┐
 │   Spatial Domain Analysis       │
 │   (Extracting LSB Bit-Planes)   │ ──► LSB Detection
 └─────────────────────────────────┘
                  │
       (FFT / DCT / DWT Transform)
                  │
                  ▼
 ┌─────────────────────────────────┐
 │  Frequency Domain Manipulation  │
 │  * Target High-Frequency Spikes │ ──► FFT/DCT/DWT Detection
 │  * Apply Low-Pass / Masking     │
 └─────────────────────────────────┘
                  │
       (Inverse Transformation)
                  │
                  ▼
      [ Sanitized Clean Image ]

🏗️ Tech Stack

• Backend Algorithms: OpenCV (cv2), Numpy, PyWavelets, Scipy
• Frontend Engine: CustomTkinter (Tkinter Wrapper)
• Packaging: PyInstaller + PIL (Dynamic Logo Generation)