CV Dazzle - Computer Vision Camouflage
A complex, but increasingly important concept, computer vision (CV) is the collection of algorithms that devices ranging from security cameras to Facebook's PhotoTagger use to automatically detect and recognize human faces. CV Dazzle provides a camouflage from computer vision by distinguishing irregular facial modifications that disrupt the mechanized, Big Brotheresque process. A process that is seen by many--including Dazzle's creator, Adam Harvey--as a violation of privacy, and infringement on American civil rights.
The Dazzle moniker was derived from a camouflaging technique of the same name used to mask warships' directionality, size, and orientation during WWI. Combining various tested methods of makeup application, hair styling, and other temporary, neck-up modifications, Harvey has established some general principles for thwarting CV efforts that do not require the use of more suspicious--and sometimes illegal--incognito accoutrements, such as masks and hoods. The beauty, and irony, of Harvey's research is that it ventures to hide its subjects by bedazzling them in bold patterns and colors, rather than covering them up, or blending them in. In addition, while the camouflage eludes artificial intelligence, it typically would not alter the wearer to the point of being unrecognizable to other human beings.
In the test subject photos, solid green lines indicate positive facial detections, while computer vision equipment deemed images with no green squares face-free. The designs used in the evading looks were inspired by tribal paint and London club scene aesthetics, though when actual photos from both were tested, many of the more outlandish visages were unable to fool detection algorithms.
CV disruption requires mastery of a few basic techniques, which can then be applied according to personal tastes. For example, you should not wear feature-enhancing makeup, such as eye shadows or liners, but instead focus on darkening or obscuring areas that typically appear light. The bridge of your nose. Your upper cheeks. Partially obscuring your eyes with strands of hair is also effective, as ocular position and coloration are key identifiers. Check out the video to see a suckered PhotoTagger unable to recognize one of Harvey's test head shots as a human face. CV Dazzle protocol can also outwit Google’s Picasa and eblearn.
Obviously, decking oneself out in tribal warpaint and rocker 'dos to avoid recognition at the airport or on Facebook is neither practical, nor a priority, for the average American citizen. However, given the radical shifts in Federal laws, exponential increases in identity theft, and worldwide access to just about anything you post about yourself--or someone else posts about you--online, CV Dazzle's quest to help us move through life unpegged and unnoticed may be an important step in maintaining our rights to privacy and anonymity during what is certain to be a progressively more invasive future.
CV Dazzle is an open-source project. Harvey hopes to post his code, which is currently written in Java/Processing using OpenCV1, to github in the near future. Check his website for updates.