How Image Recognition Can Improve Brand Insights Into Consumer Behavior

Brands can now learn more about their customers via images than ever before.

Earlier this month we documented a new startup named Curalate that offers image analytics that is able to determine whether a post on social media will gain likes, comments and shares. Now their investors are on the hunt for a slicker model.

Apu Gupta and his co-founder and CTO Nick Shiftan started with an idea that would enable brands to get involved with Pinterest in a more calculated, meaningful way by offering a service that would analyze Pins for their potential to expose campaigns and products. As they¬†began to build the business, they¬†realized many people don’t use text on Pinterest as they do on other social media sites and therefore needed to figure out a way to specifically analyze images on their own.

1.jpg

The team alongside newcomer Louis Kratz, a machine vision expert, researched alternative machine learning techniques such as multi-index hashing and the Discrete Cosine Transform algorithm. These techniques enabled the system to cluster similar images and sort large numbers of photos (millions per day), into groups and then quickly determine which photos were identical.

2.png

They have now successfully introduced a service that uses their image-recognition strengths to help marketers at companies like the GAP and Urban Outfitters learn more about how their customers’ use of images of their products on a number of social networks including Instagram.

The ways in which this service can help are huge, for example as most companies have multiple images of the same product, this idea can help find which specific image is more popular. This type of service is available in-house, however what Curalate offers is a way for brands to monitor the huge amount of activity on social media, which is where consumers are their true selves.

Now companies have a way to figure out why consumers like the brand or product as opposed to simply what they like.

Curalate

Quantcast