This post was originally posted on the WITNESS blog. It was written by Bryan Nunez, Tech Manager at WITNESS.
Last month, YouTube launched a feature that allows blurring on videos uploaded to their site. It's a step we've pushed for from the commercial video-sharing platforms and social networks – as a way to enable easy, faster, more accessible options for preserving and enabling visual anonymity in a networked, visual age. You can learn more about why we think its important in our earlier blog post and online video chat.
In this post we want to analyze what we've seen of the tool so far, and provide some tips on usage.
The YouTube facial blur feature is a step in the right direction, especially considering the way we used to do it. In terms of ease of use, it's a tremendous time saver; however it does have its limitations.
Processing large video files with complex images, lots of movement and multiple faces can take quite a bit of time. In this video of a protest in Egypt from last year, dozens of people walk through the frame and with the exception of a few faces towards the end, virtually every face is blurred. The time it took to blur out the faces in this video (about one minute long) was approximately half an hour from the time the initial upload completed.
In this second example, we wanted to test the face blurring accuracy in low light conditions. Again it does very well, ignoring faces wearing masks, but catching almost everyone else. The video is 16 seconds long and took about 5 minutes to process.
Interviews with one person speaking make up a large percentage of human rights videos (and are often the most high-risk scenario for sharing) as well as videos produced by groups we work with. So we wanted to test high bit rate HD videos with a single interviewee. As you can see the blur function is totally accurate if not a bit over zealous in it's blurring of the person's face. In fact, it actually obscures the subtitles in video, so we would need to redo them using Amara or YouTube's closed captioning feature. This video, about one minute long also took about 5 minutes to process.
Finally, I wanted to put the YouTube face blur in a head to head showdown with ObscuraCam, a mobile app for Android in development by the Guardian Project with WITNESS.
Using YouTube's automatic face blur tool:
Using ObscuraCam full automatic face detection and blur:
Although hardly extensive, these initial tests show that the YouTube facial blur feature can fairly quickly and accurately blur out the majority of visible faces that appear in any video you upload. More testing still needs to be done with respect to processing speeds relative to file size, image complexity, and data rate. YouTube also states that they have made it ‘incredibly difficult’ to de-blur images – however, we are not able to independently confirm this yet. In future releases I would like to see support for some type of audio obfuscation as well as the ability to manually define regions for blurring.
Compared to ObscuraCam, YouTube's facial blur has a much larger diameter which easily covers most detected faces, while ObscuraCam's smaller region can sometimes miss the edge of a face if it's moving. In future versions, we will add the ability to change the size of the area to blur.
Here are some tips, advisories and warnings you might want to consider with any type of blurring or anonymizing tool or when filming in an insecure context. We also provide some particular tips related to the YouTube tool. We've put them in roughly the order of how you might think of them in a video production workflow: from before shooting through to uploading and preservation (archiving).
How-to Guides: A number have been shared online, here are some good ones:
Always start with assessing risk: Starting with a risk assessment is key to all video work. Review the tips in the WITNESS ‘Before Filming’ video to consider steps you should take before you even take out your mobile phone or camera.
Aim to secure informed consent: Since there can be no absolute guarantees on anonymity, WITNESS encourages human rights defenders, including citizen activists to think about how meaningful informed consent is secured if they are filming or interviewing vulnerable people. This practice of informed consent is particularly important when filming interviews with people in private (rather than when filming protest activity in the street). We find that giving people a sense of the worst case scenario (for example, that someone's testimony will be seen by local security forces responsible for abuses) enables people who are speaking out to better make decisions on whether to be filmed, and on what level of anonymity they need.
Think about other aspects to security: Other risks to your safety and security, and that of people you film may come from how you shoot and transmit footage – for example on a mobile device. Good sources of information on other aspects of information security include the SaferMobile project of MobileActive, the ONORobot project of the Tactical Technology Collective, and the tools and materials of Frontline Defenders.
Solutions on your own device or before you upload may be preferable: Remember that your footage is vulnerable to seizure at many points from when you create it to when you share it. If you are concerned that your video might be seized before you can upload it – have a think about using a tool that is on your own device (for example, WITNESS and the Guardian Project's own ObscuraCam app for Android phones), or using less high-tech obscuring techniques. You can backlight someone for a basic obscuring technique, or not film their face: we provide some tips for this in the video below.
The YouTube blur function blurs big: If you're filming an interview best to film a waist-up shot rather than a close-up unless you want the whole screen to be blurred. This is particularly important if you're thinking of putting subtitles on your video.
Blurring does not guarantee protection: Remember that you can be identified from many different elements in a video. Some of these are visible – for example, an identifiable location you are in, or a distinctive item of clothing you wear. Some of them are heard – you may have a distinctive voice or accent. And you can also be identified by context and people you're seen with. Beyond this, remember that the username of the uploader or their IP address might be trackable, and via the person who uploaded the video people seen in the video might be identified.
No tech solution is perfect: Some face blurring approaches are reversible. Although YouTube has stated that they have made it extremely hard for someone to reverse their blur, there are no guarantees that can be made on this front.
PRESERVATION and ARCHIVING
Holding onto your original footage may put you at risk: YouTube's default option on the blurring feature is to delete your originals. Consider whether you want to do this and remember that YouTube complies with lawful requests from law enforcement including subpoenas, so they might be required to hand over your original footage if they receive such a request. As YouTube spokesperson Jessica Mason notes in a response to a blog on Slate:
With this tool we are giving users the option to delete the original “unblurred” version. In the case that they choose to delete it, the video would be permanently gone from our serves and we would be unable to hand it over. This was designed with the security of activists in mind. We also take down videos that do not comply with our Community Guidelines.
In the case that a user did not delete the video and the government of one of the countries where we are launched had a valid court order or subpoena, as a law-abiding company we would have to comply.
We'll share additional tips as we learn them, but do share what you learn as you use the YouTube tool about where it works well, and where not.
Got any tips from your own experience using the face blur feature on YouTube or other techniques for keeping subjects safe in your videos? Please share them with us.