Group Photo Finder
High-precision facial detection and matching across large, unstructured photo collections.
Context & Problem
Sorting through hundreds of high-resolution group photographs after events to find pictures of specific individuals is laborious and impractical with manual human filtering.
Solution & Approach
Developed a multi-stage computer vision pipeline combining deep learning facial detection with 128-dimensional embedding classification. The system extracts faces from group images, generates feature vectors, and matches target individuals with high precision.
Key Features & Highlights
- Robust multi-face detection handling varying lighting conditions, scales, and partial occlusions.
- 128-dimensional facial embedding extraction for scale and rotation invariant comparison.
- High-speed vector similarity matching across large image datasets.
- Achieved 90.73% identification accuracy evaluated across diverse real-world event datasets.
Results & Impact
Automated event photo indexing and search, achieving 90.73% accuracy on multi-person benchmark test sets.