With our colour asset search you can have any colour you'd like... and it does not have to be black!

Posted: 18 July 2013
Good search technologies are taken for granted in any digital asset management system worth the name. But there’s still much that can be done with the content itself to make it more discoverable.

With pictures, it’s not just metadata describing the literal meaning of what it shows that makes for a richly referenced – and therefore useful – asset. Other properties, like the predominant colour scheme, can be an essential criterion in finding and selecting the right asset for the job.

Our Colour Asset Search

While the human eye is pretty good at grouping colours, and recognising subtle differences that help us judge whether things match or clash, computers are unsurprisingly more exacting (and exhaustive) in the way they manage colours – individually identifying 14,923,237 of them for starters. We’ve done some research and developed a solution that brings these two disparate approaches to colour management together, and as a result offers our clients a powerful additional tool for organising and accessing their brand assets.

Our approach has been to enable the automatic classification of images by their overall colour using three simple measures, and further to allow searches within a range of related colours as well as from a range of similarly coloured, but darker or lighter pictures.

Colour Asset Search made simple with a Colour Mean

We start the process by establishing the colour mean – this means that we analyse the image histogram – a graph that represents all the colours in an image.

01

By chopping off either end of this graph, we iron out any extremes and arrive at an average colour for the image: the colour mean exact. This still leaves us with that unwieldy choice to make between 14,923,237 colours, so we need to map the exact number into a more general colour space (known as the Lab), that groups together similar colours, irrespective of how bright or dark they are, and by doing so reduces our options to a few thousand. The Lab colour space enables us to do this by storing information about how light or dark a particular colour is separately from the colour data itself, and so gives us the colour mean simplified.

02

Simple here is only a relative term, though, and a long way from a friendly colour asset search interface. To boil it down some more, we attach colour mean names to the various points of the compass that can be read of from the Lab space colour wheel. And because we’ve stripped out all the light and dark data, all our original millions of colours can be read off in this way.

We go with, Black, Blue, Brown, Gray, Green, Orange, Pink, Purple, Red, Sky Blue, White and Yellow.

You’ll notice that we have added some additional names to the list beyond those the colour wheel supplies. This is because human beings have a richer and more descriptive way of talking about colour, and we’ve tried to add at least some of that linguistic detail back in. But the basic principle is the same, we map the colour name back to the colour that corresponds to its position on the colour wheel.

These examples show that our basic approach is extremely effective, and how, through an intuitive Brand Centre® interface, it gives another dimension of discoverability (and therefore additional value) to our clients’ assets.

03 

(e=exact mean, s=simplified mean and the last values is the calculated colour name.)