When a resident photographs a street, a festival or a blocked pavement, the subject of the picture is rarely the only thing in it. Faces, licence plates, house numbers and other identifying details arrive in the frame uninvited — often unnoticed by the person taking the shot. Any system that asks citizens to contribute photographs inherits that problem, and cloud-side redaction only moves it: the unredacted image still has to be uploaded first.
GANonymizer removes sensitive information from photographs automatically, using deep learning, and does it on the device. A contributor takes a picture; the tool identifies and removes the sensitive content before the image is shared anywhere.
Running at the edge rather than in the cloud is the substantive design decision. It means the version of the image containing faces or plates is never transmitted, never stored remotely and never depends on the security of a server the contributor cannot see. Privacy protection becomes a property of the handset rather than a promise made by a platform.
– Automatic detection and removal of sensitive content in photographs
– Deep learning based, requiring no manual redaction by the contributor
– Runs on the device, before sharing or upload
– Unredacted image never leaves the phone
– Designed to pair with participatory sensing applications
Used in M-Sec Use Case 4 in Fujisawa, Japan, alongside SmileCityReport, so that residents could report on city events by photograph without surrendering the incidental private detail those photographs contain. The approach was carried through all three participatory phases, including the cross-border exchange with Santander.
Get in touch to explore deployment opportunities or propose a complementary approach.
