Explore this resource too.
Netherlands-based ANP uses artificial intelligence and facial recognition to tag photos faster and more accurately. Client searches now return more relevant images, increasing revenues while reducing operational costs.
Founded in 1934, The Algemeen Nederlands Persbureau BV (ANP) is the largest news agency in the Netherlands. Over 90 years it has survived many waves of disruption, not least the digital transformation of the past three decades. Now, the organisation is poised to take advantage of artificial intelligence and machine learning which promise to revolutionise journalism and news media.
Photo-journalism and image licensing is one such activity where ANP has an enormous stake. It manages a digital archive of more than 100 million images, which includes photographs from its press agency partners in France (AFP) and Germany (DPA - also a Mobius client). Every day it adds a further 50,000 images to its database including up to 2,000 from its own team of photographers.
In order for each image to be found efficiently, it must be tagged so that it matches the search intent of ANP’s clients. These are typically other media outlets looking to illustrate their own articles with professional quality photographs.
Tagging presents a number of challenges, as Patrick Rasenberg, Product Manager Photo, ANP, explains. “Until recently we had one person working full time tagging photographs as they entered the system. Other editorial staff and photographers also contributed. Even so, given the sheer volume of images there was typically a lag of anything between 24 and 72 hours as they attempted to keep up.”
The number of accurate tags is also vital. Too few will limit the exposure of an image, but too many imprecise tags may render a search meaningless. “As an end-user I want my search to return photos that are relevant to my needs. If my screen is filled with pictures that bear little relation to my article, it’s very easy for me to switch to a competitor or simply decide to run the story without a picture,” says Patrick.
The challenge is even greater because most clients are familiar with world-leading search engines and e-commerce websites which lead the way in end-user search experience. “We can’t just benchmark against our competitors, we have to adapt to the broader expectation of our clients,” says Patrick.
The commercial pressures are equally daunting. Gone are the days of thousand-dollar invoices for the exclusive rights to a single photograph. Instead, photo agencies are moving to a high volume, low margin, royalty-free model. True, there is greater demand thanks to an explosion in the number of digital publishers, but any business which fails to connect their content with this burgeoning marketplace faces a perilous future.
The situation came to a head in 2018 when ANP was sold to Talpa Network. Shortly afterwards, ANP acquired its largest Dutch competitor, photo agency Hollandse Hoogte, doubling the number of photos in the ANP archive. At the same time, competitors and agencies in other regions were beginning to deploy artificial intelligence-based solutions for the automatic tagging of photographs, although in many cases this was expensively developed, in-house software.
Patrick and his team held conversations with a number of image recognition specialists. “Some of these were massive players in the field of search, others were more media focused, but specialised in video.” ANP also reached out to Mobius Labs, curious to discover if a younger, Berlin-based computer-vision specialist held the answers.
“We were looking for two things,” says Patrick. “First, we wanted any technology to recognise the emotional content of a photograph, not just physical objects. For example, a photograph of a woman practising yoga. Can the system add abstract tags such as ‘happiness’, ‘calm’, or ‘mindfulness’?”
Secondly the team at ANP wanted to work with a partner that understood the photo agency business. “This is a traditional sector whose digital learning curve has been incredibly steep in the past few years. We need to work with people who understand the forces that shape our sector today and into the future.”
Face recognition was also a priority. To evaluate this feature, Patrick sent Mobius Labs and a second organisation images to train their machine learning models. Once this step was complete, the ANP team picked another 300 images for classification. Mobius Labs achieved a 94% success rate that astonished and impressed the agency in equal measure. Patrick says, “Mobius met our two main objectives: highest success rate, including emotional recognition while demonstrating a clear understanding of our industry. Together they contributed to our decision to select their computer vision technology.”
Unobtrusive and easy to use, the Mobius Labs solution runs in the background adding keywords and names (facial recognition) automatically. “End users including clients and editorial staff, won’t even know it’s there,” says Patrick.
Mobius Labs computer vision algorithms can tag photographs in milliseconds, using keywording and facial recognition that can recognise over 5,000 objects, emotions and actions. With more precise and consistent tagging, ANP expects to see an increase in the number of photos found and purchased by clients.
Even a single figure percentage increase in volume will deliver a business case for the deployment. “The speed and accuracy of the Mobius SDK is excellent. It will ensure that we squeeze as much revenue as possible from our archive and new additions to the collection,” says Patrick.
Editors will spend far less time on data-entry and more time making critical editorial decisions. “Our journalists can just be journalists again,” says Patrick . “It’s a good example of where a new technology makes the workplace more human and a rewarding place to work.”
Photo agencies are being squeezed by new digital competitors including free-to-download picture archives. AI-powered computer vision enables ANP to add value to its main client offer, while attracting new paying customers. Sport is one area where Mobius Labs will give ANP an advantage. “You can train the software to recognise all the players in the Dutch football league as well as events on the pitch: a header, shot, save, tackle or a goal celebration. This saves an enormous amount of time and contributes to the commercial potential of an image,” says Patrick.
The lightweight Mobius Labs computer vision algorithms are deployed as part of an SDK on ANP’s technology platform. This means that ANP remains in possession of all its data, easing compliance with GDPR and other data privacy legislation. At the same time, ANP enjoys the freedom to customise the Mobius Labs solution in step with its own business and software roadmaps, and changes in the wider competitive landscape.
ANP expects that computer vision technology will play a role in the battle against fake news, propaganda and ‘deep fake’ content. “Our new mission statement is ‘Start With the Facts’,” says Patrick. “Our clients place absolute faith in the integrity of our image archive and the accuracy of the new software will ensure that we preserve and build on that trust.”
Looking to the future, Patrick hopes to expand the relationship with Mobius Labs. This includes an ‘aesthetic ranking feature’ which measures the quality of new photographs. “All of our photographers are excellent professionals. But in today’s 24-hour news cycle, they are under greater pressure than ever to deliver to unforgiving deadlines. With less time to reflect on their captures, we hope that the aesthetic ranking feature will help them filter their best images based on composition, depth of field, position of subject, contrast and so on.”