By Sarah Saunders
Ten years ago, the very first IPTC Photo Metadata Conference in Florence was packed with photographers and picture libraries eager to discuss ways of protecting their work in the digital environment. The image industry has expanded enormously since. The image industry has the additional challenge of vast numbers of images crowding the web, and the difficulty of finding the relevant picture, as well as the metadata relating to it.
IPTC Photo Metadata Conference 2017 was designed to look into the future, with a focus on image search using AI or Artificial Intelligence. The ability of computer systems to learn from humans has increased enormously in recent years, and the necessary computer power has now become available. The question for the conference was – how far can these systems help the professional image industry sharpen up its search capability and make gains in productivity?
Solution for Auto Tagging in the Image Industry
Kai-Uwe Barthel, professor of visual computing at HTW Berlin University gave a clear exposition on the history of the field and of the pressing need to create solutions for the picture industry. There are now too many images for classical search systems to handle, but using Neural Network Analysis – a variant of AI – computer systems can be taught to tag and recognise images in a fraction of the time it takes to manually tag. As most online images are untagged, a combination of human tagging and visual similarity search presents a viable way forward. But Barthel and his team have also researched new methods of presenting images, using three dimensional structures to dig into results with large numbers of images visible at one time.
Speakers on the use of AI came from across the industry, presenting solutions which can be put into practice now. The key to success in this area is to have enough content for the computers to learn from, and this can be achieved in a number of different ways. General AI systems produce good results for skies and beaches and general themes because they’ve had the data to learn from. But users can set a system to learn from their own content so that more specialist content can be tagged if the conditions are right. Computers are learning faster, and need fewer images to learn from than before.
AI systems can be trained to recognise faces, text, colours, composition, scenes, and objects, but can also be trained in the aesthetics of image selection, with one speaker maintaining that twenty images are enough to train a system in a particular brand aesthetic. But speakers admitted also that defining the precise location of what is shown in an image by its content was tested but it did not work in a reliable way.
Speakers stressed that the important element in computer learning is the understanding of the nature of the material to be tagged, an attribute which is currently not about to be taken over by artificial intelligence. The benefits in speed and productivity will be enormous but we’re not yet talking about doing away with human skills altogether!
IPTC Video Metadata and Easier Cross Media Distribution
The first afternoon session by IPTC Managing Director Michael Steidl was about the IPTC Video Metadata Hub (VMH), published in 2016 to provide a standard set of fields for use across the varied technologies used in video. Many of the Video Hub fields are equivalent to those in IPTC Photo Metadata Standard, which helps streamline cross media distribution. The VMH can be applied down to the level of video clips, which makes it a useful metadata tool for production, archive and distribution.
Technology That Protects Rights Information in Google Image Search
The IPTC conference was held a day after a CEPIC seminar on Google. The Google image search scrapes images from their original sites and displays them in its own environment. This is bad for rightsholders as images can be saved and downloaded direct from Google without reference to the original site. Picture libraries and agencies lose significant traffic to their sites as a result with a German agencies survey indicating a drop of 50 percent in traffic. The recent fine levied by the EU on Google for anti-competitiveness in comparative shopping sites is encouraging and has paved the way for scrutiny of Google’s actions with online images.
The second presentation of the afternoon presented a solution for the problems raised in the Google seminar. SmartFrame technology allows images to be presented online without the danger of being scraped by Google as this is disabled by technical means. Most of the mechanisms people use to download images – like right-click – are disabled too. Images can be shared as links so social media sharing doesn’t lead to an image becoming orphaned and lost in the websphere. And when an image is viewed as a thumbnail in Google, there is a clear indication that it is a copyrighted image, and a link back to the originating site. Rob Sewel, Pixelrights CEO demonstrated how product items within an image could be linked back to a brand website, providing ways of funding photography in the future where photography provides a link to a paid-for advertising service. The technology could be put to all sorts of uses in both commercial and non-commercial fields, and gives control back to creators and their agents.
The success of this kind of technology, as with all solutions to image grabbing and orphaned images, lies in the uptake of the technology. To be truly protective of copyright, client websites would need to implement a technology like SmartFrame.
The IPTC Photo Metadata Conference 2017 was fascinating from start to finish for the about 60 attendees on location, the level of presentations was extremely high, and the presentations and videos are all available on the website at https://iptc.org/events/photo-metadata-conference-2017/.
Sarah Saunders runs Electric Lane, an independent DAM consultancy specialising in workflow planning, asset retrieval, data management and DAM project management. She works with IPTC’s Photo Metadata Working Group.