Foundations of Facial Recognition
summary
Distills detection vs recognition, shows a live pipeline: collect and clean faces, split ~70/30 for training/testing, extract ~60–90 nodal features into vectors, reduce 784‑pixel inputs with PCA/iigen faces, then match against stored embeddings. Demos use OpenCV and a local LBPH recognizer with ~64 grayscale snaps and Yale faces; JavaScript + WebGL hits ~49 ms vs ~3 s plain JS (claimed up to 1105× speedup). Notes practical errors: Amazon Rekognition false‑matches 28 congresspeople and higher bias/error on darker female faces.
AI-written summary · about
chapters (9)
- 0:00Introduction to facial detection
- 1:05Machine learning process overview
- 10:34Facial recognition pipeline
- 20:14Classifiers and feature detection
- 29:12Dimensionality reduction demo
- 37:26Live facial recognition demo
- 46:05Tools and libraries overview
- 51:46Performance and JavaScript benefits
- 53:30Ethical concerns and limitations
related talks
40m"Introduction to Face Processing with Computer Vision|"
31mFace Recognition
52mFace Recognition: From Scratch To Hatch
26mSelf-Portraits in JavaScript
26mNew Face, Who Dis? Recent Adversarial Approaches to Facial Recognition
21mBuilding your own dystopic surveillance state with Python
49mReal Time Face Detection through a Webcam Using AI
33mReal-Time Face Recognition: Training Fast and Accurate Models