DAM and image recognition: we can finally take it seriously

  • 26-Jan-2017

RSG subscribers have asked me for years about the viability of automated image recognition. Until very recently, I more or less told people that it wasn't worth considering as an option. But in the last year, the capabilities of such technologies — while still no replacement for an human subject matter expert — have improved dramatically.

In 2016, DAM vendors began to seriously experiment with machine/deep learning algorithms. Facial recognition was the starting point, quickly followed by the ability to identify other objects within the image itself: people, animals, plants, things. Once identified and verified (by a human initially), the software learns more of the content and starts to create relevant tags and keywords based on that content; this can then be auto-assigned directly to the asset and embedded if necessary.

Microsoft Project Oxford, Google Cloud Vision, Amazon Rekognition, IBM Visual Recognition Software, and Clarifai Image and Video recognition are all gaining traction within the DAM industry, and are getting OEM'd by vendors to analyze images upon ingest, usually by way of a cloud-based API. Thus, the learning and image identification capabilities are not limited to your own corpus of images. As these new learning algorithms improve over time, they can become a valuable resource for companies struggling with the deluge of metadata needed for success.

We expect this capability to grow in the coming years while these neural nets expand beyond what objects can be recognized in an image to the ability to identify emotions (happy, sad, angry, etc.) and sentiment. In turn, they’ll deliver contextually relevant images as part of the consumer experience. Automated, personalized, and dynamic distribution of assets to all channels will be aided by artificial intelligence, machine learning, and automation bots based on metadata that’s provided or generated. Because these functions are still very experimental, third-party relationships with providers may change over time along with pricing structures. Expect some vendors to have long-lasting partnerships and others to chop and change based on how emerging APIs (and pricing models) pan out.

This level of automation may seem anathema to the creative process, and indeed it will never replace the design and artistic intelligence that sets any creative asset apart. Creative genius cannot be measured or replicated by software, but the management — and especially the analysis — of the content and lifecycle (from creation, curation, through to consumer feedback) is an enabler of deeper creative and emotional connections.

As I mentioned above, automated tagging is still no replacement for subject matter expertise embedded as metadata at the asset level. Both vendors and you, the customer, are on a long journey using both bots and human interaction to curate metadata. Although these systems offer clear benefits, we all need to approach the notion of outsourcing human work to a robot with a proper degree of skepticism.

In our DAM evaluations, we analyze the image recognition capabilities of several vendors. As a next step, you can download a complimentary DAM research vendor sample.


Our customers say...

"Thank you for bringing something like the Digital Asset Management Research to the market. It's worth every penny."

Andy Niemann, Photo/Computer Arts, Royal BC Museum, Victoria

Other Digital Asset Management posts

Make the Right DAM Decisions in 2019

  • December 11, 2018

DAM will remain a hot topic of conversation for us in 2019. We'll be tracking developments and sharing our observations throughout....

DAM Scenario 7: Photo and Asset Library

  • November 28, 2018

In essence, the photo and asset library scenario sets forth the idea of a single, definitive catalog to determine what the organization owns....