Walk into any organization — I am sure in 100 percent of them — and somewhere you will find a hard drive full of images about which no one knows. These are files with less-than-illuminating names such as 1.jpg, 2.jpg, and so on. Someone in the past was in a hurry and did not take the time to name the files properly or to tag them with keywords. Without tags, these images cannot be searched, making this collection a veritable black hole of dead images. If you cannot find them, you cannot use them.
Machine Learning to the Rescue
I do not blame people for skipping on metadata. Properly tagging metadata can be tedious without the right planning or tools. But it still has to be done. Without metadata or tags, our images, videos, and rich media can be lost, ending up as useless bits on a disk.
Now we can use machine-learning algorithms to help with the onerous task of adding metadata to images. Image-recognition algorithms are getting better all the time, and in a recent report, Microsoft reported an error rate slightly lower than that of humans for a large set of test images.
These reports have some people panicking that machines are taking over, but it is not like that at all. At an airport recently, I checked in on my phone, completed a security check at a computer, checked my bags by myself, badged in, and did not see anybody until I got to the plane. I did everything by machine. Machines take care of certain jobs so we can do more important ones. Far from being a job killer, we now can perform strategic and high-level tasks instead of standing there checking bags.
It is the same in marketing. In a future world, machine learning can aid the process, but it does not replace full metadata management by a skilled human librarian. Humans definitely have a place: some things that will stump a computer — such as identifying cars by year, make, and model — are still obvious to people. BUT computers will not complain about having to add metadata, they will not try to avoid it, and they will work just as hard on the 100,000th image as on the first one.
Content Intelligence Can Add Value
If we run that hard drive of mystery images through a machine algorithm, they become discoverable. Marketers can find and use them again, eliminating the need to create or buy more. The images have really come back to life. This is instant return on investment.
Adobe is leveraging this technology. Smart Tags (Beta) was showcased as a sneak peek at last year’s Summit and will be released at this year’s Adobe Summit; this innovation leverages deep-learning technology to tag images automatically with useful metadata, allowing easy image discovery by marketers.
The technology can automatically tag images with keywords based on photo type (macro, portrait, etc.), popular activities (running, skying, hiking), certain emotions (smiling, crying), popular objects (car, road, people), animals, popular locations, primary colors, and more.
Machines can help with images that are already in your digital asset-management system. These algorithms may add tags, not originally thought of, to previously tagged images to make them more discoverable, increasing their value across the organization.
Another area of value is Asset Insights: unique to Adobe Experience Manager, customers can now gain immediate insights into how specific assets are performing and use that information to optimize their investments in creative that drives marketing campaigns.