Image processing goes mainstream

The saying ‘a picture paints a thousand words’ is believed to stem from an article by Fred R. Barnard in the advertising trade journal Printers’ Ink. The article, published in 1921, promoted the use of images in advertising on the side of trams. And ever since, images have been an integral part of advertising, using their power to great effect.

But in the same way as Web 2.0 has transformed the old, one-way communication of advertisers to consumers into a dialogue between brands and customers, the use of images in communications is no longer exclusive to advertisers and marketers. On the contrary, images have become a natural element of everyday conversations, since smartphones allow us to create and share images in a blink.

It might feel like the overwhelming majority of the imagery people share on Facebook and Instagram depicts either babies, pets or food – or any permutation of these three – but this view is probably biased by your life stage. In reality, people share almost every possible day-to-day experience, from holiday and shopping trips, to boiler breakdowns and crowded trains.

And they use images either to complement their posts, or to share just an image and maybe a couple of hashtags to let the picture mainly speak for itself. So to mine and analyse this part of social data, we need not only the ability to analyse text in huge quantities, but also ways to computationally analyse, categorise and score large volumes of images.

But fear not, the tools to do exactly this exist. The rise of cloud computing and machine learning has recently seen vast leaps in image processing. Computer vision algorithms have reached unprecedented levels of accuracy in face recognition, image tagging and even emotion recognition. There are multiple open source projects out there which provide access to cutting-edge academic research developments for every developer.

Check out this curated list of current open source projects on github for example. And there is an increasing number of cloud-based solutions that provide access to image processing APIs (Application Programming Interfaces) for relatively little money, such as Clarifai.com, Alchemyapi.com, or IBM Watson’s image analytics suite.

So the question is not how to do image processing, because APIs and open source projects enable anyone with some basic knowledge of Python or Javascript to create databases of millions of images including tags and maybe even emotion scores. The question is: what will be the next killer application of image recognition to create valuable insights for our clients?

To date, face and emotion recognition within market research is used mainly in the area of concept and ad testing. Researchers wish to infer the effectiveness of an ad from respondents’ supposed emotional reactions measured via micro-expressions. Even if you believe that consumers – numbed by the daily advertising bombardment – will grimace in joy or disgust into their webcams while watching a 20 second ad, you have to admit, it is not innovative to use new technology to do the same thing as before, just because it is hip to claim ‘we measure real emotions’. 

So it is up to us researchers to create smart use cases and applications for image processing, in order to extract valuable, insightful information from this essential element of human communication. Because if we don’t, technology companies will be more than happy to bite off yet another piece of the research budget cake.

Frank Hedler is director advanced analytics at Simpson Carpenter