JF-D / Face-Recognition

this code uses LBP and KNN to do face recognition.

Face Recognition with LBP

this code uses LBP and KNN to do face recognition.

Details

1. Local patch
   • P = 8, R = 1
2. mapped patterns
   • according to the 0-1/1-0 hop counts to map the initial 256 patterns to 58 uniform and 1 non-uniform patterns.
3. classification
   • use KNN (try different k) method to do classification.
   • use Histogram intersection to evaluate the distance of two LBP descriptors.

Run

1. make sure the input images' path is right.
   • in this dir, image path is CroppedYale/ + path in recognition_*.txt.
2. train.m extracts train images' LBP descriptors.
   • save them in a file.
3. run run.m
   • the acc will be given.
   • this code takes quite a long time to do face recognition.

Result

in test set:

k	acc
50	0.9641
15	0.9716
5	0.9679
1	0.9868

choose k = 1.