The document proposes a semantic context-aware privacy model called FaceBlock that uses semantic web technologies to dynamically infer user preferences about having their photo taken based on contextual information like location, time, activity, and relationships. It describes how FaceBlock works by exchanging identity and face identifier information between users' devices, recognizing the context, and triggering privacy policies to determine if a photo should be allowed or obscured. Some challenges discussed include face recognition, context and policy management, and ensuring user privacy is maintained.
Understanding Discord NSFW Servers A Guide for Responsible Users.pdf
A Semantic Context-aware Privacy Model for FaceBlock
1. A Semantic Context-Aware
Privacy Model for
FaceBlock
Primal Pappachan, Roberto Yus, Prajit Kumar
Das, Tim Finin, Eduardo Mena, and Anupam
Joshi
http://face-block.me
10. All in or nothing?
A person’s preferences would depend on her
context (e.g., time, place, or activity)
Examples
“I am okay being photographed by people
I know at a private event”
“I do not like to be photographed when I
am at public places”.
11. Semantic Web
Technologies
Understand the semantics of concepts
such as “public place”, “people I know”
or “private event”
Semantically represent privacy policies
based on concepts
Dynamically infer user preferences
about pictures based on context
12. Context-Aware
“[...] any information that can be used to
characterize the situation of an entity” Dey
and Abowd
13. Privacy Policies
For expressing user
preference on pictures
Constraints based on user
context model
Semantic Web Rule
Language (SWRL)
15. Example Policy
“do not allow my social network colleagues
group (identity context) to take pictures of
me (identity context) at parties (activity
context) held on weekends (time context) at the beach
house (location context)”
16. Glass User
Wishes to take pictures at the party
Runs FaceBlock in the background
Receives face identifiers and policies
Detects, recognizes and obscures the
faces as necessary
17. Others
Wishes to protect his privacy at the party
Generates face identifier
Specifies context constraints using rules
Runs FaceBlock in the background
18. Protocol
Exchange Identity
Share Face Identifier
I:
L: At T:
Beach
Colleague
House
Context Recognition
A: Party
Weekend
Policy Triggered
PrimalID, FaceBlock: True
20. Challenges
Image
Face Recognition / Detection /
Identifier Generation
Communication
Malicious Policies
21. Challenges
Context and Policy
Imprecise context
Policies - Generation, Conflict Resolution, Validity
General
Privacy Loss
Enforcement or Incentivizing
Energy Cost
22. Take aways
Users are defenseless against loss of privacy in
pictures
Novel approach for taking privacy-aware
pictures
Semantic Web technologies makes FaceBlock
smarter
Proof-of-concept implementation
http://face-block.me Thank you NSF and SWSA
Notes de l'éditeur
Lot of cameras
and the result - me being tweet/retweeted on twitter (social networks) without my knowledge
With new technology such as google glass, people are becoming more paranoid about technology
In this famous tv show, the presenter is taking on Google Glass as a privacy nightmare
Non-Technological Solution
Technical Solution - but not a practical approach, people can still take pictures
Takes a picture of the user and generates a mathematical representation of the face which is called the eigen face
Sends this information and policy which is don't take pictures using P2P networks such as bluetooth or WiFi
Uses face detection, face recognition to identify faces of user in picture and obscure them
An all-or-nothing model does not help in many real-life situations
User preferences- who is taking the picture and with whom it may be shared
An all-or-nothing model does not help in many real-life situations
Ontologies and reasoner can be useful for making the privacy model higher granularity and better control
An entity is a person, place, or object that is considered relevant to the interaction between a user and application, including the user and applications themselves.
We used a simple ontology for our implementation involving the use cases we were looking at and is based on the definition by Dey and Abowd.
An user preference on whether his face should be included in the picture or not
(Safe) SWRL rule for expressing policies
Usage of ontologies enables FaceBlock to apply privacy policy for specialization of general concepts. For example, if a student has specified that she does not want her pictures to be taken at the University buildings, it is assumed that she does not want any pictures to be taken at the University library unless specified by policy that she does not mind pictures being taken at the library.
Definition of context Synthesizer - Example policies which would be activity dependent are: “don’t allow my picture when I’m dancing” (shared by a user), “don’t allow my picture during meetings” (shared by a meeting room). The later policy will be applied to different types of meetings defined in the ontology (e.g., business meeting, research meeting).
Identity - Used for identifying users on first contact
Unique User ID (MAC ID, Social Network Ontologies)
SWRL rules - to model whether a user is allowed to take picture or not of another one we use the data property FaceBlockPictures(Person,xsd:boolean).
FaceBlock Google Glass user
FaceBlock Smartphone user
(P) Context Recognition
(P) Triggering Context Constraint and Sending Policy
(R) Receiving Policy and Taking picture
(R) Detecting, Recognizing Face and blurring it in the picture
FaceBlock is not only for users, its also for location and activities
Location broadcasting policies - e.g.: church, museum etc. (tourists)
Events broadcasting policies - for example a confidential presentation
Photographers wishing to cut off from unnecessary interferences
False positives, People not looking directly into the camera, bad quality pictures
Masquerading as someone else
Energy Cost - as we are running the reasoner, face recognition, context extraction on mobile device, energy is an issue.
Smarter - Fine grained representation of user context, Inferring implicit knowledge based on explicit facts
Put face block website url/ twitter/ g+
Put affiliations in the last slide
Affiliations
Contact information
Definition of context Synthesizer - Example policies which would be activity dependent are: “don’t allow my picture when I’m dancing” (shared by a user), “don’t allow my picture during meetings” (shared by a meeting room). The later policy will be applied to different types of meetings defined in the ontology (e.g., business meeting, research meeting).
Usage of ontologies enables FaceBlock to apply privacy policy for specialization of general concepts. For example, if a student has specified that she does not want her pictures to be taken at the University buildings, it is assumed that she does not want any pictures to be taken at the University library unless specified by policy that she does not mind pictures being taken at the library.