Aller Media Finds Success with KeyCore and AWS

Aller Media publishes newspapers and magazines across the Nordic region with a weekly circulation of 3.2 million copies. Established in 1837, Aller is also a leader in online publishing in the Nordics. Aller Media needed a way to reliably and economically tag the photos it commissioned and clear a backlog of millions of poorly tagged images. AWS Partner, KeyCore, worked with Aller’s imaging expert to build an automated image tagging tool based on existing AWS services that uses facial recognition, going from concept to proof-of-concept (POC) in 10 months.

Traditional publishers have faced many challenges as the world goes digital. Aller Media, the leading magazine and newspaper publisher in the Nordic region, has been on its own transformation journey.

The firm’s total weekly circulation of 3.2 million copies is overseen from headquarters in Copenhagen, Denmark, with the assistance of regional offices. It publishes multiple titles, including Elle, Cafe, and Familie Journalen. Its assets—including millions of photos in digital libraries—were scattered across countries and publications and difficult for others in the company to access and reuse.

Like all legacy publishers, Aller Media faced the challenge of serving readers moving to online platforms to consume content. While it has embraced that shift and publishes large amounts of online content, maintaining pace and scale without compromising its standards has been a challenge.

That’s why it decided to go cloud-based with its operations in 2015 and to simultaneously introduce a digital asset management system (DAM). The company knew it had to embrace this new technology to give it the agility it needed.

“We’re definitely expanding in the digital area. We see a pressure on our core business, and print and circulation in general, to transform. Establishing a new business is essential and we’ve got to be supported by the right tech for that journey,” explains Thomas Culmsee, Business Transformation Manager at Aller Media. That’s when Aller found it had a problem.

When Aller implemented WoodWing Assets as its DAM in 2015 and consolidated its digital assets from all operations across the Nordic region, it found it had millions of photos that were poorly or incorrectly tagged.

This meant that these assets the company paid to acquire were difficult to access and use. Misspelled names and missing or incomplete tags meant journalists and editors had a difficult time finding appropriate images.

“At that time, we had a whole department doing nothing but meta-tagging and organizing the whole archive. It wasn’t possible for them to get through this and make this searchable for end users. We needed to look into some digital technology to enable that,” says Culmsee.

The cost of tagging each photo was approximately €3.85 per image. With millions of images to tag, the time and cost of doing so manually was not sustainable. Aller had a limited technical team and needed help to build a proof of concept, and following that, a minimum viable product that would make a difference in the business.

As it had already shifted editorial operations to run in the cloud on AWS, and had implemented its DAM in the cloud, it naturally approached AWS to find a solution to this challenge.

AWS referred Aller to an AWS Partner, KeyCore, to implement a solution. What the two companies found was that this solution was going to require considerable work to make it happen.

Tests of existing systems by Aller and AWS showed that existing solutions were not able to satisfactorily recognize Nordic celebrities.

“Aller needed something to improve image sorting and tagging but nothing out there could do it. However, drawing on their expertise in image-handling and our capabilities to customize the application, they got what they needed. When each partner draws on their strengths, you can find a solution,” says Mikael Brøgger, Sales Manager at KeyCore.

Using Amazon Rekognition and Amazon Comprehend to tag photos, a custom solution was developed. It drew upon Aller Media’s expertise in imaging and image-handling and KeyCore’s development of customized code, built on top of Amazon Rekognition.

The result is a solution that generates metadata for the DAM and can easily be expanded to other representational state transfer (REST) interfaces, allowing for future evolution. The solution operates as an unsupervised, self-learning face clustering algorithm using a serverless version of DBScan accessing images loaded to Amazon Simple Storage Service (Amazon S3) by the DAM.

The results have been impressive. From 2019 to July 2020, the system has processed five million images and tagged 10 million faces. Of those 10 million tagged faces, AI discarded three million due to poor quality, such as the position or size of the subject in the image. Any edge cases can be flagged for human adjudication, but the amount of manual intervention involved has been vastly reduced. The cost of tagging these images has also fallen from €3.85 to just €0.12 per image.

For end users, such as journalists or editors who need to find appropriate photos, there is no interruption or delay. The tagging and classification all happen seamlessly. An image is uploaded and is then ready to use across the organization from any location. Aller was even able to outsource 55 percent of its design and layout because its systems allowed users to work together regardless of location.

Culmsee says he was surprised by some of the things Aller learned during the project. First among those was the off-the-shelf solution that didn’t work for Aller. It couldn’t recognize Nordic celebrities and would even do things like misidentify a picture of a pizza as a person. Culmsee also saw the power each partner brought to the collaboration. From planning to POC, the project took just 10 weeks.

It worked so well, in fact, that Aller and KeyCore are discussing the possibility of offering their customized solution to other organizations using a SaaS model. Culmsee says that many other publishers he knows around Europe are still heavily invested in manual tagging and could benefit from this solution.

“Having the right skills on both sides, with our image expertise and KeyCore’s ability to configure the application, it worked,” Culmsee said. “It was worth the investment.”