mediaeval 2015 - privacy protection filter using stegoscrambling in video surveillance
TRANSCRIPT
Privacy Protection Filter using StegoScrambling in Video Surveillance
Natacha Ruchaud and Jean-Luc DugelaySpeaker : Natacha Ruchaud
Privacy Filters in Video Surveillance
PRIVACY
PLEASENTNESS
INTELLIGIBILITY
2/15
Main objectives :
No more possibility to recognize people
Trade-of
Good looking
Visual quality, still recognize events: people walking, running, fighting, stealing….
Privacy Filters in Video Surveillance
Time consuming
Reversibility
Flexibility
3/15
But also :
Videos surveillance in real time
Re identification in case of trouble
No need to adapt parameters
Main idea
●Scrambling to protect data●Steganography to hide the protected data in an edge map which preserves global shape and boundary appearance
●Near lossless Reversibility
4/15
Main idea5/15
Edge mapScrambled RoI
Original RoI
StegoScrambled RoI
Crop ConvexHull
Scrambling6/15
Pixel can be represented in 8 bitsRemove the 2 Least significant bits of
the RoIAn XOR between the 6 bits of the RoI
and the random numbers generated with a pseudo-random number controlled by a seed.
Edge map7/15
Edge map8/15
DilatationKmeans ClusteringOriginal RoI
Edges detection AND
Steganography9/15
The 6 bits of the XOR image (pixels intensity is either 0 or 63) are shifted in the LSBs
The 2 MSBs of the edge map (pixels intensity is either 0 or 192) are added in the 2 MSBs of the XOR image
Recovered image10/15
Flowchart11/15
Results12/15
Results13/15
Future works & Perspectives
Less favorable outcome for intelligibility and pleasantness●Edges detector and Kmeans clustering distortions●Black and white results
Future works:
●Improve edges detection●New edge map
14/15
THANK YOU
15/15