Face Identification from Obfuscated Images in Deep Learning using Feature Compensation

MR. K. Perumal1 Assistant professor,
Department of CSE, AnnamacharyaInstitute of
Technology and sciences, Tirupati517520,A.P, India perumalinfo@gmail.com
K Bhanu Prakash4
UG Student, Department of CSE,
Annamacharya Institute ofTechnology and sciences, Tirupati
517520,A.P, India.bhanuprakash02062001@gmail.com
M Gowthami2
UG Student, Department of CSE,Annamacharya Institute of Technology
and sciences, Tirupati-517520,A.P,India. mallegowthami8@gmail.com
k Kavya3
UG Student,Department of CSE,Annamacharya Institute of Technology and
sciences, Tirupati-517520,AP, India.Kunanikavya906@ gmail.com
B TharunKumar Reddy5
UG StudentDepartment of CSE, Annamacharya Institute ofTechnology and sciences,
Tirupati-517520,AP,India tharunkumarreddy12.06@gmail.com

The technology of face recognition has had immense positive changes in aspects like the security of the people and convenience to the user. But the fact that it is widely used has also generated serious issues on pri- vacy since it is quite simple to gather facial images and misuse them. It poses a big dilemma as, although the requirement of the use of a reliable face recognition system is high, people are becoming more and more unwilling to provide their original facial information.To deal with this problem, we suggest that we use is the PRO- Face C that is a privacy-sensitive face recognition system, able to maintain the required balance between privacy and accuracy. The suggested approach is based on the client-server approach in which obfuscated face image is sent over the server only by the client of the solution. Identity detection is then done with a pre- trained model along with privacy- free complementary features so that original face appearance is never actually obtained, yet the obfuscated images still provide good preview of an image and a mechanism of identity-guided feature compensation mechanism is used to further increase the accuracy of recog- nition. Moreover, there are a number of privacy-related measures, which are implemented to reinforce data security even more. A comprehensive test on various face recognition data sets indicates that the proposed system has very high recognition performance at minimal user privacy at different contexts.Index Terms—Face recognition, privacy preservation, obfus- cated images, client–server architecture, feature compensation