a summary of use case face recognition (what, how, real life application and risks)

1. A case of Deep Learning :
Face Recognition
BY : HERMAN KURNADI

2. intro
Face recognition is the problem of identifying and verifying
people in a photograph by their face.
Deep learning methods are able to leverage very large
datasets of faces and learn rich and compact
representations of faces, allowing modern models to first
perform as-well and later to outperform the face recognition
capabilities of humans.
It is a task that is trivially performed by humans, even under
varying light and when faces are changed by age or
obstructed with accessories and facial hair. Nevertheless, it
is remained a challenging computer vision problem for
decades until recently

3. WHY face
recognition
(automation*,not
manually**)
•To restrict access to a resource to
one person, called face
authentication.
•To confirm that the person
matches their ID, called face
verification.
•To assign a name to a face, called
face identification.
**Humans can perform this task very easily.
*hard problem to perform automatically with software, even after 60 or more years of research.
i.e.:, recognition of face images
acquired in an outdoor
environment with changes in
illumination and/or pose remains a
largely unsolved problem. In other
words, current systems are still far
away from the capability of the
human perception system

4. • Locate one or
more faces in the
image and mark
with a bounding
box.
1.Face
Detection.
• Normalize the face
to be consistent with
the database, such
as geometry and
photometrics.
1.Face
Alignment. • features from the
face that can be
used for the
recognition task.
1.Feature
Extraction.
• Perform matching
of the face against
one or more known
faces in a prepared
database
1.Face
Recognition.

5. • uses hand-crafted filters that search for and locate faces in photographs based on a deep
knowledge of the domain.
• Can be very fast and very effective when the filters match, although they can fail dramatically
when they don’t, e.g. making them somewhat fragile.
Feature-based face detection
• learns how to automatically locate and extract faces from the entire image.
• Neural networks fit into this class of methods.
image-based face detection
• consists of a sequence of simple-to-complex face classifiers and has attracted extensive
research efforts.
• has been deployed in many commercial products such as smartphones and digital cameras.
• they often fail to detect faces from different angles, e.g. side view or partially occluded faces.
Detector (cascade)
Face Detection

6. •Face Matching:
• Find the best match for a given face.
•Face Similarity:
• Find faces that are most similar to a given face.
•Face Transformation:
• Generate new faces that are similar to a given face.
•Face Verification.
•A one-to-one mapping of a
given face against a known
identity (e.g. is this the person?).
•Face Identification.
•A one-to-many
mapping for a given
against a database of
known faces (e.g. who is
this person?).
Problem of face recognition : supervised predictive modeling task trained on samples with
inputs & outputs.

7. •based on deep convolutional neural networks
•first major leap forward using deep learning for face recognition,
•accuracy of 97.35% reducing the error of the current state of the art by more than 27%, closely approaching human-level
performance(human standard is 97.53%)
DeepFace
•much like DeepFace, although was expanded in identification and verification tasks by training via contrastive loss.
•Accuracy of 99.15%
The DeepID,
•uses a deep convolutional network trained to directly optimize the embedding itself, rather than an intermediate bottleneck layer as in
previous deep learning approaches.
•To train, we use triplets of roughly aligned matching / non-matching face patches generated using a novel online triplet mining method
FaceNet
•to collect a very large training dataset and use this to train a very deep CNN model for face recognition that allowed them to achieve then
state-of-the-art results on standard datasets.
•a very large scale dataset (2.6M images, over 2.6K people) can be assembled by a combination of automation and human in the loop
The VGGFace (Visual Geometry Group,Oxford)
Deep Learning for Face Recognition

8. Academia
•The GaussianFace algorithm developed in 2014 by researchers at Hong Kong University
• facial identification scores of 98.52% compared with the 97.53% achieved by humans.
•weaknesses regarding memory capacity required and calculation times.
Facebook
•in 2014, Facebook announced the launch of its DeepFace program which can determine whether two photographed faces belong to the same
person,
•with an accuracy rate of 97.25%.
Google
•in June 2015, Google went one better with FaceNet, a new recognition system with unrivaled scores: 100% accuracy in the reference test Labeled
Faces in The Wild, and 95% on the YouTube Faces DB.
•Using an artificial neural network and a new algorithm,
•This technology is incorporated into Google Photos and used to sort pictures and automatically tag them based on the people recognized.
•it was quickly followed by the online release of an unofficial open-source version known as OpenFace
Microsoft, IBM and Megvii
• FACE++ tools had high error rates when identifying darker-skin women compared to lighter-skin men.
Amazon
•cloud-based face recognition service named Rekognition to law enforcement agencies.
•The solution could recognize as many as 100 people in a single image and can perform face match against databases containing tens of millions of
faces.
•falsely identified 28 members of US Congress as people arrested for crimes.
Development

9. 1. Security - law enforcement
•to combat crime and terrorism
•is used when issuing identity documents, combined with other biometric technologies such as fingerprints.
•is used at border checks to compare the portrait on a digitized biometric passport with the holder's face.
•Drones combined with aerial cameras offer an interesting combination for facial recognition applied to large areas during mass events
Health
•track a patient's use of medication more accurately
•detect genetic diseases such as DiGeorge syndrome with a success rate of 96.6%
•support pain management procedures.
Marketing and retail
•KYC
•By placing cameras in retail outlets, it is now possible to analyze the behavior of shoppers and improve the customer purchase
•Since 2017, KFC, the American king of fried chicken, and Chinese retail and tech giant Alibaba, have been testing a face recognition
payment solution in Hangzhou, China
Use cases

10. China
• video surveillance network countrywide. 176 million surveillance cameras were in use at the end
of 2017 and 626 million are expected by 2020. Chinese police is working with artificial
intelligence companies such as Yitu, Megvii, SenseTime and CloudWalk
2020 Olympic Games in Tokyo (Japan).
• will be used to identify authorized persons and grant them access automatically enhancing their
experience and safety.
India,
• Aadhaar project is the largest biometric database in the world.
• It already provides a unique digital identity number to 1.2 billion residents.
• Face authentication will be available as an add-on service in fusion mode along with one more
authentication factor like fingerprint, Iris or OTP
Use cases

11. •anti-spoofing mechanisms Make sure the captured image has been done
from a person and not from a photograph (2D), a video screen (2D) or a
(3D), (liveness check or liveness detection) Make sure that facial images
(morphed portraits) of two or more individuals have not been joined into a
reference document such as a passport
1.In Russia, Grigory Bakunov has invented a solution to escape the eyes
permanently watching our movements and confuse face detection devices.
He has developed an algorithm which creates special makeup to fool the
software. However, he has chosen not to bring his product to market after
realizing how easily it could be used by criminals. 2.Forbes announced in an
article from May 2018 that researchers from the University of Toronto
have developped an algorithm to disrupt facial recognition software (aka
privacy filter).In short, a user could apply a filter that modifies specific pixels
an image before putting it on the web. These changes are imperceptible to
human eye, but are very confusing for facial recognition algorithms.
Pro Cons

12. A Gentle Introduction to Deep Learning for Face Recognition by Jason Brownlee on May 31,
2019 in Deep Learning for Computer Vision
https://www.gemalto.com/govt/biometrics/facial-recognition
Source

13. Foto Ini oleh Penulis Tidak Diketahui dilisensikan atas namaCC BY