Anonymizing Audiovisual Data


Can you help establish best practices and tools for anonymizing audiovisual data?
If so, write to 
and we will connect you with a mentor.

More information

Audiovisual data from public broadcasts does not need to be anonymized, but a range of data is in the form of private recordings of individuals. Various supervisory agencies, such as Internal Review Boards at universities, may prohibit the researchers who made the recordings from sharing them with other researchers whose projects could benefit from access to the data. Marie-Louise Brunner, Stefan Diemer and Selina Schmidt have been able to satisfy the requirement for anonymization by using the "Find Edges" filter in Adobe Premiere Pro as part of their CASE project after testing various alternatives. The filter ensures anonymity, prevents reconstruction of the filtered data and preserves facial cues and gestures. An example of the result is available hereAnother example, created by Turner from a TV broadcast, is available here Adobe Premiere Pro offers a wide range of filters and parameters, and presumably produce much more sophisticated and nuanced specific effects; the examples cited here were created using only the default settings in Adobe Premiere Pro. The process is quick and easy: drag and drop the video into the Premiere Pro window and select the "Find Edges" effect under the "Stylize" menu. Jungseock Joo explains that a similar contour & edge effect can be achieved using OpenCV, which is open source. Steen observes that this "means we can automate it and run it on a headless server." Here is an example of edges created by OpenCV:

This kind of data would of course not be made public but could be held by Red Hen to be shared with Red Hens who make a convincing case that their research would benefit from the loan of selected anonymized data files. Who can inform us about the possibility of including in proposals to supervisory agencies an explanation that the data would be anonymized and placed in Red Hen's archive? Can we anonymize retroactively, thus permitting us to ingest a vast trove of protected data that gesture researchers have analyzed? The project of anonymizing audiovisual data is related to the project of integrating ELAN tags into Red Hen.