I came, I saw, I conquered and I redacted ??? You may be wondering what am I doing by adding to an already well-known quote “Veni Vidi Vici”. Well, that's the order of the day, when we have numerous computer vision algorithms to detect, track and recognize, we are having to move towards redacting all of that and do that efficiently as well. While there are numerous applications for facial recognition and detection such as law enforcement, finding missing people, tracking people at events etc, privacy and protection become the major undercurrent behind redaction. With the introduction of GDPR which protects the privacy of the customers and makes data exchange transparent, redaction has become very relevant. For most businesses, its significance has already been recognized even before its popularity.
Redaction implies obscuring or removing something, in this case, an object, face, audio or even frames of video. Redaction basically involves detecting, tracking and blurring any face or object in any video, it sometimes also redacts audio or small clips of video frames from any long video. The first step to redaction is detection, as the object or face to be redacted needs to be detected first. Algorithms and models are employed to detect faces and objects and bounding boxes are placed on the same after detection. These bounding boxes are then blurred or pixelated depending on the requirement in redaction. In a video, to continuously redact the object or faces, they also need to be tracked every frame. Face detection and tracking as I mentioned before becomes an integral part of the redaction, however, in some cases facial recognition also becomes essential. There are some situations which require a particular person, child or object needs to be redacted from the videos, in such cases recognition comes very handy and the steps proceed as face detection, face recognition and then face redaction.
As mentioned before, privacy is the major driving factor due to which redaction was born. The prevalence of image and video capture through smartphones, body cameras, video cameras and surveillance cameras and the advancement of technology where these can be shared throughout the world has actually increased the concern for the protection and privacy. Also sometimes the request for such videos as evidence or surveillance for proving criminal activities etc has also put people or objects not involved in a particular criminal event at risk. Videos generally involve people who are the perpetrators and also innocent bystanders who are then put at risk as their identities are exposed. In some cases, non-involved license plates or sometimes even small children are there in the video who are not a part of what is being monitored. Another instance would be video monitored retail shops where customers pay with their cards and their numbers are exposed or if the person is shoplifting and only that person should be highlighted. In these cases, in order to protect those who are not involved, redaction comes in handy.
Redaction workflow typically has detection, tracking and redacting and if it is a particular person or object, then recognition also comes into play. This can be done manually or automated. Automatic redaction relies on accurate detection mechanisms being paired with robust tracking methods across the video sequence to ensure the redaction of all sensitive information. So, the next time we are under the “scanner“, remember our information is secure, intact and will be protected by redaction and protection laws