1. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico The project is co-funded by the European Union, through the Safer Internet plus programme http://ec.europa.eu/saferinternet The project is co-funded by the European Union, through the Safer Internet plus programme http://ec.europa.eu/saferinternet Content Goals Goals Team Team Partners Partners Organisation Organisation Milestones Milestones Details Details Questions QuestionsContent Goals Goals Team Team Partners Partners Organisation Organisation Milestones Milestones Details Details Questions Questions

2. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico I-DASH goals To develop a collection of integrated tools capable of handling thousands of hours of videos potentially containing child pornography. The tools will focus on filtering out irrelevant material, recognition of known child pornography, interactive search in the material and establishing links between different videos. To establish a standard to exchange video fingerprints and corresponding data to facilitate international collaboration and development of new kinds of technology. To establish a European database with video fingerprints of known child pornography, allowing for efficient filtering of data and facilitating collaboration between national police forces.

3. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico I-DASH research and development team OrganizationCountryContactMain Focus University of Amsterdam (UvA)Netherlands Marcel Worring, m.worring@uva.nl Project coordinator Concept detection Interactive search Nederlandse organisatie voor Toegepast Natuurwetenschappelijk onderzoek (TNO) Netherlands Wessel Kraaij, Wessel.kraaij@tno.nl System Architecture VideoFingerprinting Instituto Superior Technico (INESC) Portugal Isabel Trancoso, Isabel.Trancoso@inesc-id.pt Audio analysis University of Surrey (Surrey) United Kingdom Josef Kittler, J.Kittler@surrey.ac.uk Location matching Object matching ZiuZ – visual intelligence (ZiuZ)Netherlands Gerrit Baarda g.baarda@ziuz.com User requirements Dissemination Application Development System Integration

4. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico I-DASH end-users OrganizationCountry Ministère de la DéfenseFrance InterpolInternational National Police Agency The Netherlands The Netherlands PolitietNorway Child Exploitation and Online Protection Center England RikskriminalpolisenSweden

5. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico I-DASH milestones DateMilestone 3Q2008Installation VizXview and first prototype of I-DASH 4Q2008 Representative end-user selected for steering committee First version of user requirements and focus points 1Q2009Full design of the system ready 3Q2009 Installation of second prototype system and demonstrators (working on own video material) Conference 1Q2010Installation of third prototype system 3Q2010 End of the project Final conference

6. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico I-DASH Genre Classification Genre Classification Complete Captured Video Collection Recognizing known material Recognizing known material VIDEOfingerprint Potentially relevant data Potentially relevant data Harmless material Harmless material New material New material Known Child Porn Known Non-CP Known Child Porn Known Non-CP Interactive search engine Interactive search engine Videos to explore Videos to explore Interactive Search engine TV-recordings, Hollywood movies, sports, harmless home videos or normal porn Audio: screaming, fear Video: faces, bed Audio-visual: children, porn Videodatabase, known childpornography Captured Video collection Captured Video collection Captured Video collection Captured Video collection Captured tactical research investigation trainee investigator expert Location and Object matching

7. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Classify large amount of video and images Exact copy detection (MD5, SHA1) IMAGEfingerprinting Or near copy detection or Near hashing on images VIDEOfingerprinting Nearcopy detection on video Video linking Location matching Object matching Interactive search Genre classification Concept searches Off-the-shelf productsHigh science VizXview (video) VizXimage (picture) I-DASH first prototype I-DASH project

8. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Overview Baseline: VizXview and VizXimage VIDEOfingerprinting Interactive search

9. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Baseline Video: 1 hour in 1 minute Images: 1.800.000 on a single machine Nudity detection MD5/SHA1 Hashing IMAGEfingerprinting (other names for this techniques are nearhashing or nearcopy detection) Huge codecset for both video and images More then 350.000 hours of video classified so far More then 300.000.000 images classified so far

10. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Videofingerprinting Aim –Identify video material which has already been classified in earlier cases as being child pornography

11. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico For video youll often find … Resolution change Color change noise transcoding

12. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico MD5 hashes work for exact copies only: 32 characters, hexadecimal A0D5DA345 original 342350954 modified 81237987F color 023984098 noise 912837912 size 230948019 crop transcoding EDF12645A Videofingerprinting will yield a signature of each video which is robust against the various types of transformations

13. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Global workings of VIDEOfingerprinting Video database Fingerprint Repository Videofile Fingerprinting Annotation Key Adding a videofile to the repository

14. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Video database Fingerprint Repository Videofile Fingerprinting Searching Annotation Key Searching/matching for a videofile in the repository

15. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Interactive search Both from a technological viewpoint and from a business perspective there are three approaches Genre classification to support the classification process. Filter out harmless material like normal home video, Hollywood movies, sport videos, normal porn, … Concept search to support the classification process. To find videos with screaming children, children expressing fear, interviews, faces, porn, nudity, bed or … Location and object matching to support crosslinking and tactical investigation

16. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Genre Classification Potentially relevant data Captured video collection Clearly Harmless Data Genre Classification

17. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Automatically detect TV-recordings, Hollywood movies, sports, harmless home videos or normal porn Based on for example average shot length number of different camera positions movement image quality Remarks: Technology is based on statistics. The longer the video the better the prediction. Do we trigger on a hidden child porn scene of 1 minute in a sportsvideo of one hour? Genre classification

18. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Semantic search engine New material Ranking of the data based on relevance to the query Audio Indexing: screaming, children, pool Visual Indexing: indoor, person, pool, beach Concept based search

19. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Location matching Photo 2 is taken from a different angle and with a different zoom level photo 1 photo 2

20. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Automatic matching on details in the room

21. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Object matching Photo 2 has another background and another pose photo 1 photo 2

22. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico The tattoo is matching

23. University of Surrey, ZiuZ, University of Amsterdam, TNO, Instituto Superior Technico Dissemination Website: www.i-dash.eu Annual dissemination conference and presentation: june 2009 Project result presentation: june 2010 3-monthly newsletter: first november 2008 Set up of a partner program: start december 2008 Contact: dissemination manager of the project –Gerrit Baarda (ZiuZ, g.baarda@ziuz.com, +31 652616233)g.baarda@ziuz.com