The Magic of Image Analysis
The proliferation of imagery on social media, particularly Instagram, has created both an opportunity and a challenge for brands.
- The opportunity is to tap into this rich vein of content to understand what people are posting about your brand, or about your category, and of course to use your own brand’s account for marketing and engagement.
- The challenge is to be able to analyse, monitor and make sense of all of this in a meaningful and consistent way.
Over the last year or so, we’ve noticed that an increasing number of clients are coming to Signoi with a need for a fast and reliable way to decode what’s going on in terms of visual language and how it may or may not speak to their brand. Equally, we have seen growth in wide-concept briefs such as ‘what’s the visual discourse in health’, or ‘how are concepts like courage expressed on Instagram’.
And, there have been questions around the role and imagery promoted by influencers.
We’ve even been asked to use our metrics to decode online imagery to see what image characteristics can be determinants of hard measures such as clickthoughs, likes, shares, and so on.
Always the underlying theme has been how best to use imagery to spark the imagination.
That’s why we’re excited to have developed a new set of AI-driven quantitative semiotic methods for this hard-to-analyse category of stimulus. We call our new approach to image analysis Imagysis for short…
For example, taking Instagram as a source, now you can download, map, measure, analyse and compare the visual language and comments:
- Within a category (say, soft drinks). What does spontaneous UGC on Instagram tell you about the semiotics of the category?
- From your brand’s and competitors’ accounts (say, @pepsi, @cocacola and so on). Understand the visual language that’s being used and what that means for a brand’s positioning as well as identifying new, emergent and white space semiotic opportunities to move the needle.
- From the wide open world of hashtags (say #tropicana, #perrier, #innocent etc). Learn from how people are contextualising your brand within their everyday lives.
To bring this to life a little, here is an example of a small-scale test project, in the soft drinks category.
Here, we downloaded the several thousand Instagram posts from the @accounts of a number of well known brands, including Pepsi, Coke, Evian, Perrier, Tropicana, and more. With a sample of several thousand images and associated text in hand, we deployed our AI driven platform to categorise and cluster these, using a variety of in-house metrics representing the content, form, colour, structure and inherent meaning of the images.
This reduces the dataset to a number of core themes that tend to repeat in consistent patterns. For example, ‘vitality’ and ‘purity’. A small selection of imagery from the hundreds of images clustered into each theme is shown below.
You can immediately see the innate similarity of the images in each of the clusters – they are not all the same by any means, but there is a consistency of tone and meaning that short cuts our interpretation in an intuitively human way.
From these various dimensions, the Signoi platform creates a map to represent the territory. A two dimensional representation (of what is a many-dimensional space) is shown below. All the clusters are distinct, but some are further away from others. The map represents the ‘visual space’ occupied by the images. It goes without saying that every project produces a different structure and a different map – it all depends what you feed in.
The analysis models each high-dimensional object (i.e. the image) by a two- or three-dimensional point in such a way that similar objects are nearby and dissimilar objects are more distant with as high a probability as it can within the bounds of a 2-D approximation. So points that are near to each other are by definition expected to be similar in content and thematic meaning.
Finally, because this is essentially a form of quantitative semiotics, we can go on to measure the ‘visual DNA’ of each brand. The bar chart below tells us what proportion of each brand’s Instagram feed is made up of each visual cluster.
So you can see Lucozade Sport majors on the visual language of Energy. Real Kombucha plays to Purity, Innocent to Simplicity, Dirty Lemon to Vitality – and so on. Whilst most of the themes/clusters are present for most of the brands, the ‘visual DNA’ of each is subtly or not-so-subtly different.
All this creates a means of mapping the category – and each brand’s position in it – according to inbuild visual language. But more than that, it enables brands to have a measurable view of the thematic content of the imagery they are using. It can play out into advertising too…
In sum, the quantitative semiotic engine behind the Imagysis method can deal with tens of thousands of images, and delivers as standard:
- Clusters of meaning, highlighting a manageable number of core themes within the imagery, as well as subthemes that may embody emergent codes.
- Archetype analysis and Universal Energies, showing what the imagery is transmitting in terms of characteristics and human motivations.
- Territory maps, showing how themes relate to each other within a category, and how brands position within this.
And of course, this is just a small taster of what Imagysis can do.
So when would you use Imagysis and why?
- Category understanding – a powerful additional lens into the visual language of your sector.
- Consumer understanding – finding out what implicit needs are represented by the visuals people post – like a large scale projective technique.
- Brand understanding – measuring what you are transmitting and what it likely means to the recipients.
- Marketing effectiveness – determining what types of imagery and visual language generate most engagement.
- Competitor understanding – compare, contrast, and act upon the visual DNA of any and all brands within your category.
Get in touch today to learn more about Imagysis and what this powerful new approach can do for you.
Please contact us at firstname.lastname@example.org