Elastic Bunch Graph Matching (EBGM) | The Algorithm That Saw Faces Differently

Before deep learning, how could a machine recognize faces without powerful GPUs or millions of training images? In 1997, Elastic Bunch Graph Matching (EBGM) topped the FERET benchmark by representing a face as an elastic graph of Gabor-wavelet features instead of raw pixels. This video visualizes the full algorithm with Manim. ⏱️ TIMESTAMPS 00:00 - Intro 00:07 - The problem: recognizing a face on the street 00:33 - Why face recognition is hard (intra-class variance) 01:02 - Before deep learning: the world of 1997 01:59 - The core idea: three pillars of EBGM 02:35 - Gabor wavelets: filters inspired by vision 03:12 - The jet: a "texture DNA" at one point 03:41 - Comparing two jets (magnitude & phase) 04:17 - Image Graph: a face as a graph 04:41 - Face Bunch Graph: stacking many faces 05:06 - Similarity between two graphs 05:33 - Elastic Graph Matching: stretching to fit 06:19 - The recognition stage 06:42 - Results on the FERET database 07:19 - The role of phase & computational speed 08:03 - The big picture 08:39 - Conclusion 🎓 COURSE CSC14006 — Pattern Recognition · Individual project: algorithm analysis & visualization with Manim Faculty of Information Technology, University of Science, VNU-HCM 👤 STUDENT Bang My Linh · 23122009 · [email protected] #FaceRecognition #EBGM #PatternRecognition #Manim #ComputerVision #MachineLearning #fit #hcmus #ai #ml