Facial De-Identification

About

With the advance of camera technology and face biometrics, many organisations and governments are exploiting the potential of data sharing to enhance their operations or business models. However, this has inevitably ignited concerns about the privacy of the individual(s) identifiable in the recordings. De-identification in multimedia content can be defined as the process of concealing the identities of individuals captured in a given set of data (images, video, audio, text), for the purpose of protecting their privacy. Scientists at the University of Hertfordshire have developed a unique face de-identification system that enables the options of swapping an original face with either an existing face image of user’s choice (e.g. the face of a celebrity, a cartoon character, or an animal) or a face that is automatically synthesized by the UH system. Unlike other technologies currently available, the UH system: Produces a distinguishable face for each individual while retaining the same diversity as the original face set within a corresponding de-identified face set. This means faces that look similar before de-identification will remain similar afterwards and faces that look different will remain different; and Reduces the re-identification risk of the de-identified (synthesized) faces to nearly zero. This novel development opens many application opportunities in the field of on-line personal privacy protection. The UH system works fine with both still images and live videos. Technology Overview: Masking, blurring and pixilation are the ad hoc face de-identification methods that were widely used in the past. However, these methods fail to serve their purpose as they are either reversible or unable to thwart the existing face recognition software. Moreover, they are all highly destructive and hence destroy data utility. The University of Hertfordshire scientists have developed solutions using the k-Same family process, where sets of images are partitioned into clusters identified in size by the value term “k”. The k-Same system had a number of limitations which the team have successfully resolved. It has been proved theoretically that the UH face de-identification method guarantees a near zero re-identification risk regardless of the value of k. The near zero re-identification risk has been confirmed by the testing results against some most advanced face recognition methods including LBP (local binary patterns), HOG (histogram of oriented gradients ok?), and LPQ (local phase quantization). The face de-identification system produces a distinguishable face for each individual while retaining the same diversity as the original face set within the de-identified face set (i.e. faces that look similar before de-identification will remain similar afterwards and faces that look different will remain different). Model-based FET (facial expression transfer) has been proposed and incorporated within the UH face de-identification system to restore the facial movements and expressions in the original images/videos. The system completes the face de-identification process by blending a de-identified face region with its original background, where the new shape of the de-identified face region is retained without distorting other objects in the image and the skin tone within the face region is adjusted to achieve a seamless blending with the other parts of the human body (e.g. ears, neck). The University of Hertfordshire has a prototype ready from demonstration which runs in real time. Further Details The intensive adaptation of surveillance has inevitably ignited concerns about the privacy of the individual(s) identifiable in the recordings. This growing concern along with the associated ethical and social responsibilities has made de-identification become the focus of attention by many organizations. The European COST framework is hosting Action IC1206 ‘De-identification for Privacy Protection in Multimedia Content’ (2013-2017). The US National Institute of Standards and Technology has published its first guidelines on ‘De-Identification of Personal Information’ last October. JPEG has launched a new activity entitled ‘Privacy and Security’ last September. The goal of de-identification is twofold: 1) removal of original identity to protect privacy; as well as 2) preserve the naturalness and usability of data to facilitate further data analysis and exploitation. The Data Protection Act 1998 of the UK, the 2015 General Data Protection Regulation of the EU, and the Consumer Privacy Bill of Rights proposed by the US government in 2015 all demand the deployment of appropriate technical and organizational measures to protect private information in the course of transferring or processing such data. This legal requirement along with ethical responsibilities has restricted data sharing and utilisation while various organisations may require the use of such data for research, business, academic, security and many other purposes. Potential Applications / Potential Markets Privacy protected public data sharing and dissemination, for example TV interview under witness protection - replace an existing face with a synthesized face through face de-identification More options for online video / photo sharing of personal data – people with access permission can see the original image/video while people without access permissions see a de-identified face with the original facial movements and expressions (rather than be refused access) Film, game industry - facial expression transfer between the faces of different characters or between different versions (e.g. between photo version and cartoon version) of the same face Beauty industry - to demonstrate the impact on a client’s face by, for example, a cosmetic surgery, age, weight loss, makeup, etc. with a dynamic video (with the expressions of the client) rather than a still image. Face swapping for fun – replace an existing face with another existing face of a celebrity, a cartoon character, or an animal. The University is seeking partnerships for technical collaboration or licensing opportunities.   If you would like to speak to Enterprise Europe Network to discuss the University prior to contacting them please contact: Nicky Whiting – Innovation Advisor [email protected] +44(0)7921353734