Online Reputation Management by Brand Image Recognition – Sysomos Gaze
While Online Reputation Monitoring is well established in textual social channels the missing link – and it’s a big one – has been the ability to monitor images for brand occurrences. Through a sophisticated process of digital brand decomposition and reconstruction this ability is now available – to monitor all brand occurrences within images.
A prize worth fighting for
Facebook has tightened access to individual posts and images but Instagram, with 400 million active monthly users, 40 billion total photos shared, 80 million photos uploaded per day and 3.5 billion daily photo likes, is an image treasure chest. And then there is Twitter with about 500m tweets per day – and applying an estimate that about 35% have images attached – that’s 175m images per day being uploaded.
The question is – how can you track your brand in all those images? Or even better how do you track your brand and your competitors brands in those images? Ever since Google re-ignited the deep learning approach to image recognition, famously through the Google Brain project which demonstrated automatic recognition of cats in Youtube videos, image recognition has moved ahead at a rapid pace.
Say hullo to KINSHIP’s newest partner Sysomos Gaze. Gazemetrix was recently acquired by Sysomos, and KINSHIP is Sysomos’ leading partner in Asia Pacific.
Brand tracking in images – NAB
We set Gaze up to find images which contained the NAB logo. Of the 2191 Instagram photos during the last 7 days which contained NAB in hashtags, comments or images 20 were verified as containing NAB logos. This may seem like a small number, raising the possibility of some being missed. However if you examine the photo below you will see just how accurate the image recognition process is.
In this night scene in Hobart the NAB logo is barely visible in the top left, but the Gaze algorithm has verified the NAB logo.
In the image below the logo is more easily verified than in the previous image.
If we examine the accompanying text we can see this this was a protest against NAB, apparently mounted in a busy street. The Gaze brand matching algorithm is able to cope remarkably well with the angles of the logos and the varied lighting and contrast conditions.
Finally, in the image below we can see how the crumpled NAB logo on the boy’s shirt (on the left) is identified correctly by the system.
Use cases for image brand recognition
There are numerous value-added use cases for this type of automatic brand recognition:
Firstly and obviously maintaining brand reputation – company best practice;
Ensuring that the brand is not used in ways which reduce brand reputation;
Monitoring activism against the brand – as in the Audi image above;
Measuring marketing effectiveness as in NAB and the AFL;
Measuring engagement at different events and locations (using the Gaze geolocation filter) e.g. do we find that certain locations and events get much greater image sharing with our brand visible?
Identifying brand advocates through images alone;
Undertaking competitor analysis of image engagement and visual brand reach;
Doing competitor comparisons of image engagement and sharing at different events to understand which events best capture image engagement i.e. not “share of voice” but “share of imagery”;
Directly measuring effectiveness of marketing campaigns which promote images containing the brand.
Given the newness of the technology the limits are not yet known, but suffice to say that just starting with some simple use cases will be a significant step forward in both reputation management and marketing research and effectiveness analysis.