What started as a side project for one of the developers at Griffeye is now a highly appreciated and requested new feature in Griffeye Analyze DI Pro. A month ago Face Recognition for images was introduced – and in the latest release it is now also possible to recognize faces in video material.
Humans are highly skilled at identifying and recognizing faces, better even than technology. But the opposite is true as soon as we start dealing with large volumes of data. Where there are millions of images and thousands of hours of film it is almost impossible for humans to pick out and correlate faces. This is where technology comes to the rescue. Technology can process a week of surveillance video or ten thousand video clips, and break out all unique individuals, thereby narrowing down and structuring the material. Investigators can then prioritize the information and look through the relevant material manually.
Griffeye Analyze DI has long had Face Detection, which makes it possible to detect faces in images. With Face Recognition, it is not only possible to detect faces, but also to identify and connect them to find all similar faces in a case. This possibility has now been extended to video.
In the first scenario, the investigator needs to locate and identify unknown individuals (suspects or victims) in a case. They can use Face Recognition to quickly find and match all images where different individuals are depicted.
In the second scenario, the investigator has an image of a known suspect, victim or someone else that is important to an investigation, and wants to find out if this individual can be found in other material or in another case. With Face Recognition, it is possible to take that photo and search through the data to find similar faces.
Applying this to video is the next step – and a highly important one. It is particularly useful in investigations where the subjects are consciously trying to hide their faces. To solve the case, you might need a single frame where a person shows his or her face. With Face Recognition, finding that frame is possible, but imagine how often that frame would otherwise be missed – and has been missed before!
Face recognition has been around for many years. However, it was just recently that the algorithms and the ways to apply them became good enough to cope with the challenges of today’s uncontrolled digital media analysis. For example, face recognition has worked reasonably well on clear portrait headshots, but has had trouble on more unclear or complex images. The new generation of algorithms has solved this and can handle rotated images, moving environments and faces seen from the side. It is one of these high-quality algorithms that we have applied and made available in Analyze DI Pro.
The video and image Face Detection is done automatically at import and after that Face Recognition can be run in a number of different ways. It is also possible find and search for specific suspects or victims and search between images and videos.
As mentioned, Face Recognition started as a side project for one of the Griffeye developers. Released in May 2018 it was very well received. The challenge of further developing the plugin to support video was then taken on by developers at the Griffeye hackathon held in May. Now it is released. Enjoy – and let us know what you think.