EMRAYS AI Visual Context Search Tool


Regular readers will recall that a few months ago, I reviewed MerlinOne’s NOMAD (NO MetAData) search tool which allows images to be found using text keywords searches without entering descriptive metadata.

Recently, I had the opportunity to check out an AI search tool developed by a Dutch firm called EMRAYS.  Unlike MerlinOne, they don’t make their own DAM, but offer their product as an independent component that can be integrated with other systems.  My understanding is they currently have connectors with Bynder and also WoodWing’s Swivle but are also keen to partner up with other vendors in the DAM market.  At its core, EMRAYS uses a combination of sentiment data collected from 4.5 billion data points (sourced from social media sources) and a highly sophisticated similarity detection algorithm.  One or more images can then be used as ‘seeds’ to rank all other images according to whether they are positive/negative and how similar they are.

Our business idea is to help companies, institutions and governments to put people’s emotions at the heart of business.  To help to reinforce positive behavior in the world, we first need to understand the world. We’ve already analyzed over 4.5 billion emotional reactions to find the patterns that influence our lives. At EMRAYS we call this Calculating Emotions.” [Read More]

This method has some interesting applications, for example, a user can provide an existing set of assets and EMRAYS will find other images which the user has missed via a conventional keyword or metadata search.  The results can also be filtered and combined with more conventional metadata.  There are some videos on their site which demonstrate a few use cases.

One aspect of this form of search which might not be immediately obvious is its use as a tool to assist designers and other creatives to locate assets which they may not have found otherwise.  Conventional DAM search requires the digital asset cataloguer to have very clear ideas about how to contextualise assets in order to maximise their potential value.  A careful balance is needed; if keywords are too broad, the assets becomes useless for most searcher’s purposes; if they are too narrow then they are rarely found and therefore, never used.  Effective asset cataloguing is very difficult (a fact often not appreciated by many DAM users until they are forced to deal with the issue themselves).  It is quite easy to get this calculation wrong.  Many times when I am asked to troubleshoot allegedly faulty DAMs, it is the user-supplied metadata which is the issue, rather than the software which is working entirely as designed.

The use of Big Data techniques by products like EMRAYS and NOMAD can (at least in-part) help to resolve some of these issues.  The large data repositories which they can draw upon help provide the context which will help to more precisely identify what the user might be interested in finding.  They are using a semantic model like emotions, textual analysis etc. to define semantic contextual proximity of a given image to a search term.  This approach is superior to the simplistic AI pattern recognition techniques that most DAM vendors use right now (typically via a component developed by Google, Clarifai, Amazon etc.)  This is not to say pattern matching, will never have a role, but using data interpreted by human beings at a very large scale is a far better way to focus on the searcher’s intent than generically analysing pixels in an image and then locating the closest matching object.  Because the semantic context is given more emphasis in-terms of weighting, it is more likely to be representative of what the searcher is looking for.

While all this is very interesting stuff, EMRAYS is not the ‘holy grail’ AI technology which will remove the need for high quality metadata which human beings have catalogued as a result of a deep and intimate knowledge of both the subject matter and core findability concepts.  In particular, the issue of dealing with subject-specific metadata remains one which is very complex to automate.  Further, I cannot see how it will ever be economically achieved since it still requires a custom model to be developed for nearly every single subject domain where a DAM solution might be used.  With that said, EMRAYS still has a number of very intriguing uses and this is not to say progress will never be made with subject-specific metadata by them (or others) and it is a technology which we plan to keep track of.

There is a wider point to be made here about the DAM software market, in general.  It appears that over the last decade, DAM vendors have become too preoccupied with keeping their core platforms fed, watered and integrating with other solutions to have any time left to innovate themselves.  Even the larger operators (and in DAM, most of them aren’t really that large anyway) are under-resourced and/or unable to hire sufficient talent at the scale required.  Add in the fact that many firms depend on their customers for new ideas as they themselves simply have no idea where to start in terms of innovation and it is quite obvious to see why this scenario has arisen.

Although out-sourcing this kind of more cutting edge research to third parties has its risks, for many vendors, it might be the most practical approach.  In this case, product development becomes less about trying to find a few hours of development staff time to squeeze in some ‘blue sky thinking’ (and all the untested code that will accompany it) and more about researching potential partners like EMRAYS.

There is a lot of talk about the alleged growth of the DAM market and recent analyst reports have jumped on the bandwagon to claim double-digit figures are being achieved.  This doesn’t bear up with the reality on the ground, however.  The recent series of acquisitions of DAM vendors by larger CX suite providers suggests many of the former secretly know this to be true and that client revenues are being recycled rather than significant new deals inked with those who have never done DAM before.

To have growth in technology markets, it is necessary to have innovation.  If that falls away, growth usually stalls not long afterwards.  For nearly a decade now, innovation in DAM has been inexorably moving to tool providers who offer products which integrate and extend DAMs.  If there is a growth opportunity in DAM, it is probably in this segment rather than the platform vendors themselves.

Share this Article:

Leave a Reply

Your email address will not be published. Required fields are marked *