Unbeknownst to social media fanatics, they are being watched. With billions of photos shared daily on a variety of social media, a giant opportunity has opened up for brands to stealthily track how consumers interact with their products or services. Advances in content recognition has made logo detection within vast sets of images easier and with solutions on the market for the last 8 years as well as brands eager to better understand the consumption of their product, it should be a slam dunk. But it’s not, at least for the time being, and there are a few reasons why.
Not so easy.
One would think that in the realm of content recognition, logo detection should be the easiest. Usually colorful, with consistent shapes vastly different than what can be otherwise be found in nature, logos pop out. After all, that is partly what they were designed to do. While that might be true for some brand logo ( the TIDE logo), it can however be challenging for others. In fact, the simpler the logo, the hardest it is to detect. Notoriously difficult is the NIKE swoosh because of its very crude and simple shape. Not only it can be easily confused with natural forming patterns ( like shadows), it is also either mostly white or black ( very common colors) and can offer a wide array of variations, especially on clothes. Most systems fail, not by missing it but rather by returning a vast array of false positives. While machine learning is making big thrives in getting better at identifying logos in almost any situation ( ~ 89%), it is still plagued with many non-relevant findings. Often, human filtering is needed to clean up the results.
Recognition is only the beginning
Finding a logo in an image is only the beginning. Being provided with thousands of images with my logo in them is useless if I cannot make sense of it. Understanding the context is key. Is my logo associated with moments of joy or despair? Does it appear mostly in photos of groups or in solitary? Night or day? Summer or winter?
Most logo recognition companies ignore this part, besides Ditto labs. The Boston-based company has just started integrating additional content classification algorithm so that it can “see” what a brand logo is associated with in pictures. Thus, a report for a potato chip company can include how often it is seen on a beach for example. Or how frequently it is paired with a time of day and which one. From there, brands can start discovering behavioral patterns that could allow them to better adjust their marketing campaigns.
No actionable follow-up
However, the biggest challenge facing logo recognition today is the lack of actionable follow-up. Most tech companies limit their services to social media listening, offering their clients a dashboard of where and how their logo appeared and nothing else. While major brands might have the time and budget to just watch brand logo appearances, most companies do not. In order to be indispensable, brand logo recognition solutions need to offer their customers a return on their investment in the form of post analysis actions, if only suggestions. Something they currently lack.
I found my logo, now what?
In other words, what value can a brand extract from discovering where and when their logo appears and what actions can they take, based on that discovery, that would cover the cost and justify the service? Using our example above, the potato chips on the beach, how is this information I can use? Have more bags of potato chips available in stores in shoreline retail ? Change communication to show that chips are beach-friendly or instead, make a big marketing push for city consumption? And doesn’t the fact that there are more pictures of my potato chips on beaches a reflection that people just tend to take more pictures on beaches than at home? In other words, brand detection in social media can be imprecise if not deceptive.
For a brand that has spent thousands of dollars on the sponsorship of an event, logo detection is certainly an essential tool to track earned media and overall exposure. In fact, companies that sell sponsorship placement should offer it in their packages. Not only for monitoring social channels but a broadcast and print, alas, something most tech start-up companies do not offer yet.
Some companies, like Logograb, have turned the issue around and instead offer logo detection as a more advanced form of a QR code. Take a picture of a logo and you are offered various marketing. While an interesting approach, it will be a challenge to build an automated response in consumers triggering them to scan the logo if they want more info. Sysomos has just acquired Gazmetrix in order to add logo detection on its already vast array of social media monitoring tools, allowing brands to extend their reach to visual influencers for example. Gumgum, certainly to add proper in- image advertising relevant photos, has just released its stand-alone solution called Mantii. Ditto labs, while continuing to offer its social solution has lately partnered with a few existing social media mentioning companies as an add-on, Oracle Social Cloud being one of the latest. Kuznech, which we had interviewed here, also offers a stand-alone solution.
Privacy issues… do people have an adverse reaction?
While social media listening is of great interest for brands wanting to interact with its customers, it is different when dealing with photos. A text tweet, for example, is a call for comments. People fully expect a brand to react. A photo, not the same. Unless purposely documenting a defect, they are mostly illustrating a private moment to be shared with acquaintances, not with brands. If I post an image of my buddies enjoying a beer at sunset, I certainly do not expect Budweiser to jump in and comment on it. Furthermore, it would be an easy step for logo detection companies to start building a comprehensive database on who uses what brands where, allowing for a pinpoint targeting of specific consumers ( for example Coca Cola could target Pepsi drinkers). While certainly of a large appeal to marketers worldwide, it would quickly feel creepy for most users. In term, social media platforms would have to disallow such usage of their API’s in fear of turning off their users.
So what now?
Logo detection is one of these visual tech applications that look great on paper – linking the surge of visual content with marketers – but fail to properly deliver, at least for now. While there is certainly a need, current implementations deeply lack key actionable follow-ups and behavior confidence. Few brands have the luxury of time and resources to simply look at where their products appear. Furthermore, without a wider universe than just social media monitoring, they only offer part of an answer to brands, especially those investing in event sponsorship. In short, logo monitoring still has a way to develop before it can be truly useful to brands worldwide.
Author: Paul Melcher
Paul Melcher is the founder of Kaptur and Managing Director of Melcher System, a consultancy for visual technology firms. He is an entrepreneur, advisor, and consultant with a rich background in visual tech, content licensing, business strategy, and technology with more than 20 years experience in developing world-renowned photo-based companies with already two successful exits.