Why AI for Logo Detection Is the Next Marketing Must-Have
The primary trademark of any company is its logo, a simple graphical symbol that is (or should be) unique, easily recognizable and extremely useful in branding efforts. During sponsored events in particular, it is essential to assess how many people were exposed to the sponsor’s logo and how much impact this had on brand awareness.
If a company attempted to do this without the help of an automated system, it would be difficult and cumbersome — a task prone to many errors and approximations. Hopefully, this is precisely the job for AI systems, which specialize in pattern detection. Once trained on relevant images, these can offer real-time identification of logos in video footage and social media streams.
Logo Detection Goals
Since almost all purchasing decisions are subconscious, it makes sense for a company to strive to be on top of consumers’ minds at all times. Also, statistics show that a brand’s loyal audience identifies with those people who are exposed to the brand almost daily.
The amount of time a logo is featured on the screen is directly proportional to the likelihood of people recognizing it or even paying attention. Therefore, the three main goals of any logo detection software are:
● To correctly identify the chosen brand’s logo in the provided materials (images, TV footage, social media feeds, CCTV)
● To compute the total on-screen time for the given logo in videos
● To put together results from different media to evaluate the overall impact of the sponsorship campaign
How Does It Work?
The central technology which enables logo detection is called computer vision (CV). It analyzes the recorded images frame by frame to identify significant bits, in this case logos. The challenge these systems face is to determine the graphical representations on a wide array of objects of various shapes and sizes, captured under diverse lighting conditions. There is a tremendous variability regarding the support and also the use of the logo since sometimes it has slight color or tilting variations from the standard.
The system classifies the detected logos by appearance time and dimensions. As you can imagine, it is a different story if, for example, you have a pocket-sized logo versus a full t-shirt back logo.
More advanced systems can also create a heat map of each brand to see if the appearances are equally spread on the available screen space and if the larger sizes appear in the best spots.
A good logo analysis tool can identify any number of brands it was trained for, including variations of the same brand.
Training the Detector
As in the case of other machine learning applications, the system can be taught from scratch, or use some pre-trained modules, which are then calibrated.
Transfer learning, as the second method is called, is much faster and requires far less effort to label data for training purposes. Typically, a few hundred are enough, compared to a few thousand in the first case. The base of the algorithm is usually a convolutional network that is pre-trained for specific objects. The result is surrounding the object (the logo) with a border, attaching a label and an estimation that the label is correct.
Who Can Use Logo Detection?
This technology can help media and marketing agencies, TV stations and business intelligence providers to offer their clients a comprehensive report about the success of each campaign.
Marketing agencies can show their clients how the target consumers interact with the brand on different channels. In this case, logo detection software is also relevant for TV advertising companies, which can make specific offers with clear ROI and post-event performance metrics. This comes in handy for sports sponsorships in particular.
Not only marketing agencies can benefit from this technology but every company interested in measuring their logo coverage. For example, through APIs, a brand’s business intelligence system can be recalibrated to include the data from the logo detection algorithm and add the result to the measurements of brand awareness.
Sometimes the logo detection technology can help to prevent fraud and keep the brand safe from counterfeit attempts.
Detecting Logos Day In, Day Out
Not only major sporting events need to be analyzed for impact. In fact, daily uploads on social media platforms could tell more about a brand’s influence. Although on these platforms there is the possibility to use text descriptors and hashtags, most users don’t bother to do so unless they are complaining or want to show status. More casual uploads are only context-based and include the logo in a natural setting. These are the best candidates for logo detection if the company wants to assess how the product is consumed and perceived.
After a logo has been detected, the system can retrieve other metadata such as hashtags or geolocation in an attempt to create consumption maps or identify the context. Brand associations derived from such analysis are also important, as this can be the base for natural partnerships.
In 2017, a study based on logo detection strived to identify if there was a correlation between a product’s market share and the number of appearances on social media profiles, with some surprising results. It showed no direct connection between the two. For the products with the smallest market shares, the number of posts was almost as high or even higher than the products with the largest market share. A possible explanation could be the fact that since social media is an aspirational environment, people share more expensive products, which usually have a smaller market share, to enhance their social positioning among friends.
Logo detection can help companies monitor their marketing efforts and measure brand awareness, especially during sponsored events. However, it is not the best idea to scrape social media for references unless the brand is an aspirational one, which people are eager to display.
Thanks for reading The Marketing & Growth Hacking Publication
Why AI for Logo Detection Is the Next Marketing Must-Have was originally published in Marketing And Growth Hacking on Medium, where people are continuing the conversation by highlighting and responding to this story.