basicopencv-beautyfilter

Implemented basic smoothing and slimming filter. A real-time beauty filter using OpenCV's DNN face detector and bilateral smoothing.

Side project to explore computer vision fundamentals — face detection, facial masking, and selective image processing.

How it works

1. Face detection — OpenCV's res10 SSD DNN model locates faces in each frame
2. Feathered mask — an elliptical mask is built around each detected face with soft edges to avoid hard compositing lines
3. Bilateral smoothing — applied only to the face region; preserves edges (eyes, lips, brows) while smoothing skin texture
4. Skin tone correction — subtle warmth boost and redness reduction
5. Alpha compositing — the smoothed face region is blended back onto the original frame

This approach is meaningfully different from naive circular blurring — it targets skin specifically and preserves facial features.

Setup

# 1. Clone the repo
git clone https://github.com/YOUR_USERNAME/basicopencv-beautyfilter.git
cd basicopencv-beautyfilter

# 2. Install dependencies
pip install -r requirements.txt

# 3. Download model files
python download_models.py

# 4. Run
python main.py

Controls

Key	Action
q	Quit
s	Save snapshot as snapshot.jpg
r	Edit in filter.py

Project structure

opencv-beautyfilter/
├── main.py             # Webcam loop and window management
├── filter.py           # Beauty filter pipeline
├── download_models.py  # One-time model downloader
├── requirements.txt
└── models/             # DNN model files (downloaded, not committed)

Next steps

• Facial landmark detection (68-point) for feature-aware masking
• Frequency separation for texture-preserving smoothing
• Skin segmentation to exclude non-skin areas from processing
• Adapter to swap webcam input for DSLR frames (for photobooth integration)