mediaeval 2015 - mediaeval 2015 drone protect task: privacy protection in surveillance systems using...
TRANSCRIPT
MediaEval 2015 Drone Protect Task: Privacy Protection inSurveillance Systems Using False Coloring
Serdar Ciftci1, Pavel Korshunov2,Ahmet Oguz Akyuz1, Touradj Ebrahimi2
1. Middle East Technical University, Ankara, Turkey{sciftci, [email protected]}2. Ecole Polytechnique Federale de Lausanne, Switzerland{pavel.korshunov, [email protected]}
Motivation•Existing privacy protection methods do not have all of the following desirable
properties:
– Reversibility,
– Security,
– Flexibility,
– Robustness,
– Being visually non-disturbing.
•The proposed false color based method encompasses all of the desirable properties.
Proposed Method•Based on false color visualization,
•Easy to compute: Lookup tables are used,
• Secure: Lookup tables can be encrypted,
•Reversible: Lookup tables can be inverted,
•Flexible: Does not depend on format/encoding,
•Visually non-disturbing: No unpleasant artifacts,
•Adjustable: Degree of protection can be varied,
• Intelligible: Preserves non-private information,
•Robust: Does not rely on computer vision.
Privacy Protection PipelineInputImage
Gray scaledImage
Gray scale conversion
Color Mapping
False Color Result
RecoveredImage
Inverse Color Mapped Image
Converting to RGB space
Inverse Color Mapping
Figure 1: Main steps in false coloring based privacy protection.
Used Color PalettesLOCS RBS
DEF
Figure 2: LOCS, RBS, DEF color scales are selected with respect to regions’ pri-vacy level namely low, medium, and high.
Example Results
Figure 3: False color results, a) original frame, c) protected frame, e) recoveredframe.
Subjective Evaluation ResultsPrivacy Intelligibility PleasantnessFC AVG FC AVG FC AVG
Expert 0.39 0.49 0.76 0.59 0.73 0.60Naıve 0.34 0.48 0.75 0.58 0.75 0.61
Average 0.365 0.49 0.755 0.59 0.74 0.60
Table 1: Evaluation results where FC and AVG respectively stand for false colorresults and the average of all submissions. Expert and Naıve’s are participant groupswhich represent the people conducting research on visual privacy protection andnaıve observers.
Conclusions•We described a simple and effective method for protecting privacy using false col-
oring.
•Benchmarking evaluations indicated higher than average score for pleasantnessand intelligibility.
•However, the privacy protection level was found to be lower than average.
• Future work will investigate designing custom color scales to improve privacy pro-tection and the quality of reversibility while enhancing security.
References[1] Atta Badii, Pavel Korshunov, Hamid Oudi, Touradj Ebrahimi, Tomas Piatrik,
Volker Eiselein, Natacha Ruchaud, Christian Fedorczak, Jean-Luc Dugelay, andDiego Fernandez Vazquez. Overview of the MediaEval 2015 drone protect task.In MediaEval 2015 Workshop, Wurzen, Germany, September 2015.
[2] Margherita Bonetto, Pavel Korshunov, Giovanni Ramponi, and TouradjEbrahimi. Privacy in Mini-drone Based Video Surveillance. In Workshop onDe-identification for privacy protection in multimedia, 2015.
[3] Serdar Ciftci, Pavel Korshunov, Ahmet O Akyuz, and Touradj Ebrahimi. Usingfalse colors to protect visual privacy of sensitive content. In SPIE Human Visionand Electronic Imaging XX, pages 93941L–93941L–13, 2015.
AcknowledgementsThe work was conducted in the framework of FP7 NoE VideoSense and TUBITAKproject number 114E445.