Deliverable D6.1: “Design and implementation of in-browser content analysis filter”

The overall aim of the ENCASE project is to leverage the latest advances in usable security and privacy-preserving technologies, in order to design and implement a browser-based user-centric architecture for the protection of minors (age 10-18) from malicious actors in online social networks.

This deliverable D6.1 - “Design and implementation of in-browser content analysis filter” isrealized within the context of the Task T6.1 - “Automatic detection of sensitive personal content”.This Task aims to develop sophisticated content analysis algorithms able to automatically detect contents, including text, images, and videos, that may contain user private data that could eventually pose a risk to the integrity and safety of the user.

The main objectives of the aforementioned Task are:

1. a)  the generation of annotated training/tests set that would can be used for training/testing the algorithms to be developed;

2. b)  the determination of key image processing tasks in relation to the detection of private content (e.g., face recognition, expression recognition, body pose recognition, and nudity detection);

3. c)  the development of algorithms for each of the aforementioned image interpretation tasks;

4. d)  the development of the computational linguistics and NLP algorithms to analyze the content of the communication between users (messages or newsfeed posts) for evidence of

   sensitive content (e.g., address information);

5. e)  and the integration of the developed algorithms in a content analysis filter (CAF) installed

   on user's browsers, so that an on-line user is warned when content that could potentially contain private data is being uploaded from her browser on an OSN or appears on otherusers’ newsfeeds.

This document describes in detail and demonstrates the progress made towards automatically detecting this threat, with specific emphasis to the algorithms implemented to be included in the aforementioned in-browser filter.