Eyetracking, Saliency Mapping and Eye/Mouse clicks… (1/2) February 25, 2010Posted by Scott Hodgins in Advertising, eye tracking, Market Research, Marketing, Media, Studio, Technology, Tips And Tricks, Tobii, Uncategorized, Updates, Usability & UX.
Having seen the debates on LinkedIn, earlier exchanges on our blog, and Twitter it occurred to me that many who are tweeting, re-tweeting, blogging and posting their thoughts online don’t actually get the differences between these different technologies. Hopefully, this short post will help…
Saliency prediction & mapping…
What is it? These tools are, from a geek perspective very cool, although just thinking about the maths behind it will induce a major headache. The basic concept is to analyse an image and mathematically model how an average user will perceive features such as text, logos or imagery. By computing values for shape, colour, contrast, and the changes and rates of change between the features it is possible to assign values and weighting to estimate which features attract the eye of a viewer.
3M Visual Attention Service & FengGui
I believe there is a place for this, although I am not 100% certain exactly where they sit in the “long-tail” behind eyetracking. They do seem to hold some promise for very early stage analysis. As a believer in the power of eyetracking, I would, in my more optimistic moments love to see everyone eyetracking everything, every time they need to make a decision. Back in the real world this is not practical at every stage – you can’t reasonably have a room full of suitable test participants sat around on the off chance you’ll need them on any given day for a series of 5 minute tests.
So where is it useful? I can see this type of algorithm used to add weight to A/B decision making such as which version of a particular creative is likely to draw the audience – the pink or the green? Maybe the use is in trying to optimise concepts before in-depth, real world, testing. It seems like a logical step to then take this “optimised” creative and then test it in depth against a suitable target audience. Analysing an image to compute saliency, and thereby estimate what people will notice, is one thing, however, actually tracking where people are looking is the only way to truly understand how people actually interact with the stimuli they are shown.
In all, I think that these techniques could mature into a useful partner technology for eyetracking. The 3M offering seems to be the most advanced offering out there – at least in our head to head trials it’s outputs come “closest*” in a static test to eyetracking. *This last statement needs a lot of context and caveats due to the very, very different ways of working, for an honest appreciation get in touch.