Introduction Identity Data Set and Face Representation Associate-Predict model Switching Mechanism Experimental Results
Appearance-based for face recognition Inevitable obstacle Associate-Predict model The studies of brain theories
Identity data set Face representation
200 identities from the Multi-PIE data set 7 pose 4 illumination
Representation at the facial component level 12 facial components Face F = (f1, f2,..., f12) › fi for each component
Appearance-prediction model Likelihood-prediction model
Two input faces Setting : S A, S B › A and B are facial components Select the specific face image setting is equal to S B › component A’ from this image
d A = |f A ' − f B | › distance between the components d B = |f B ' − f A | Final distance between A and B: 1/2 (d A + d B )
Adaptive distance d p αA and αB : weight After the “appearance-prediction” on all 12 facial components, we can obtain a new composite face
Using classifier measure the likelihood of B belonging to A Positive training samples › Input face › the K most alike generic identities
Implement this switching mechanism › facial components : A and B › settings : S A = { P A, L A } and S B = { P B, L B } Categorize the input pair into two classes › “comparable” › “not comparable” › based on the difference of S A and S B
Comparable class › {|PA − PB| < 3 } and {|LA − LB| < 3 } Not comparable class › the rest situations
The final matching distance d sw › d a : the direct appearance matching › d p : the associate-predict model
Experiments on the Multi-PIE and LFW data sets Basic comparisons Results on benchmarks
Holistic vs. Component
Positive sample size › number of positive samples is *k › “1” is the input sample › K is the selected number of top- alike associated identities
K = 3 as the default parameter
Switching mechanism › the switch model can effectively improve the results on both benchmark
Multi-PIE benchmark LFW benchmark