US 9,811,718 B2
Method and a system for face verification
Yi Sun, Hong Kong (CN); Xiaogang Wang, Hong Kong (CN); and Xiaoou Tang, Hong Kong (CN)
Assigned to Beijing Sensetime Technology Development CO., LTD, Haidian District, Beijing (CN)
Appl. No. 15/129,567
Filed by BEIJING SENSETIME TECHNOLOGY DEVELOPMENT CO., LTD., Beijing (CN)
PCT Filed Apr. 11, 2014, PCT No. PCT/CN2014/000390
§ 371(c)(1), (2) Date Sep. 27, 2016,
PCT Pub. No. WO2015/154206, PCT Pub. Date Oct. 15, 2015.
Prior Publication US 2017/0147868 A1, May 25, 2017
Int. Cl. G06K 9/00 (2006.01); G06K 9/52 (2006.01); G06K 9/62 (2006.01); G06N 3/08 (2006.01); G06K 9/46 (2006.01)
CPC G06K 9/00281 (2013.01) [G06K 9/00288 (2013.01); G06K 9/52 (2013.01); G06K 9/6215 (2013.01); G06K 9/6269 (2013.01); G06N 3/08 (2013.01); G06K 2009/4666 (2013.01)] 13 Claims
OG exemplary drawing
 
1. An apparatus for face verification, comprising:
a memory storing one or more software applications;
a processor that executes the one or more software applications to:
extract HIFs (Hidden Identity Features) for different regions of faces by using differently trained ConvNets, wherein last hidden layer neuron activations of said ConvNets are considered as the HIFs;
concatenate the extracted HIFs of each of the faces to form a feature vector;
group the HIFs into a plurality of groups in the formed feature vector for each face, each group containing HIFs extracted by same ConvNets;
extract local features from each group of HIFs;
extract global features from the previous extracted local features; and
calculate a single face similarity score from the extracted global features so as to determine if the two feature vectors are from the same identity or not based on the calculated score.