210 likes | 345 Views
Face Recognition Committee Machine. Presented by Sunny Tang. Outline. Motivation Algorithms Review Face Recognition Committee Machine (FRCM) Experimental Results Conclusion and Future Work. Motivation. Applications in security Authentication Identification
E N D
Face Recognition Committee Machine Presented by Sunny Tang
Outline • Motivation • Algorithms Review • Face Recognition Committee Machine (FRCM) • Experimental Results • Conclusion and Future Work
Motivation • Applications in security • Authentication • Identification • Face recognition system with high accuracy
Eigenface • Application of Principal Component Analysis (PCA) • PCA: • Find eigenvectors and eigenvalues of covariance matrix C from training images Ti:
Eigenface • Face Space • Space formed by the span of eigenvectors • Training & Recognition • Project the images on face space • Compare Euclidean distance and choose the closest projection
Fisherface • Similar to Eigenface • Application of Fisher’s Linear Discriminant (FLD) • FLD: • Minimize inner-class variations and maintain between-class discriminability
Elastic Graph Matching • Based on dynamic link architecture • Extract facial feature by Gabor wavelet transform as a jet • Face is represented by a graph G consists of N nodes of jets • Compare graphs by cost function
Support Vector Machine • Look for a separating hyperplane H which separates the data with the largest margin
Support Vector Machine • Linearly non-separable data • Use of kernel function to map data into a higher dimension: • Kernel function: • Multi-class classification • “one-against-one” • “one-against-all”
FRCM • Mixture of five experts • Static structure
FRCM • Result r(i) • Individual expert’s result for test image • Confidence c(i) • How confident the expert on the result • Weight w(i) • Average performance of an expert
Result & Confidence • Eigenface, Fisherface, EGM • K nearest-neighbour classifiers • Result: • Confidence:
Result & Confidence • SVM • One-against-one approach used • For J different classes, J(J-1)/2 SVM are constructed • Confidence:
Result & Confidence • Neural network • Binary vector of size J for target representation • Result: • Class with output value closest to 1 • Confidence: • Output value
Voting Machine • Ensemble result, confidence to arrive final result • Weight w(i): • Fixed weight for each expert • Score s(i) function:
ORL Face Database 40 people 10 images/person Yale Face Database 15 people 11 images/person Experimental Database
Experimental Results • ORL Face database
Experimental Results • Yale Face Database
Conclusion and Future Work • Conclusion • Use of Committee Machine has improvement in accuracy • Future Work • Existing algorithms in FRCM do not perform satisfactorily under various lighting condition. Experts like Illumination Cone may help • Adopt dynamic structure in committee machine
Q & A Thanks!