Westminster today invited technology companies to pitch “innovative solutions” to help it identify still images used by the Islamic State or “Daesh” in its propaganda, for a competition opening on Monday 14 May.
The Home Office can determine 94 percent of Daesh videos with stunning 99.995 percent accuracy, it said, but still images are inconsistent and pose a greater challenge to identify. Some 80 percent of terrorist media is stills however.
Winning applicants will demonstrate they can “automate the detection, cataloguing and analysis of still terrorist imagery posted on a range of social networking, public messaging or content hosting platforms” the government said.
“We are looking for concepts that can be developed into workable solutions. Winning companies will be awarded up to £50,000 to develop their ideas. A further £500,000 may be available to continue development and test the successful solutions in the public sector”, the Government Digital Service and Innovate UK added in a joint release.
The competition was announced as one aspect of the first GovTech Catalyst round; a competition with five other areas. The others include more prosaic challenges like tackling waste chain issues and reducing traffic congestion.
The image recognition competition closes on 27 June 2018
Applicants can be an organisation of any size, but need to be able to handle confidential information and pass a counter-terrorism check.
The scope of the project says that applicants need to be able to demonstrate that they can input and output large volumes of content in the most efficient form, “combine metadata analysis with developing technologies to achieve desired results” and “be flexible enough to allow new models to be added, and to be easily integrated within existing detection tools.”
Speaking at the announcement of GovTech’s Catalyst challenge Prime Minister Theresa May said: “It is absolutely right that this dynamic sector [tech], which makes such an immense contribution to our economic life and to our society, has the full backing of Government.”
Image Recognition Progress
Putting in place a system to identify particular images in large blocks of data is a task being worked on by some of the biggest names in tech at the moment, primarily with commercial applications.
Recently Microsoft researchers developed a garment selection tool to help pick out cloths in images that have a lot of background noise in them.
Earlier this month meawhile Google announced a challenge to be held at a major Computer Vision conference in Munich, Germany this year. Contenders must come up with cutting edge automated image segmentation. They can work with 12.2 million bounding-box annotations for 500 categories on 1.7 million training images