Shutterstock Founder Jon Oringer MS’99 on Revolutionizing Image Search
A revolution in machine learning has empowered computers to sort and understand images without human inputs, said Jon Oringer MS’99, founder and CEO of Shutterstock, at Columbia Engineering’s annual Magill Lecture September 20.
It’s a development Oringer witnessed first-hand. After creating his company in 2003, Oringer shot tens of thousands of photos to fill market needs, manually labeling images with keywords for customers to search. Today, Shutterstock’s search engine is comprised of millions of lines of code utilizing iterative learning from over a billion queries a month to better pinpoint what users really need, even when their searches are unclear or images have few keywords.
“From day one, search was a problem, and it got more difficult as it went,” said Oringer, once named the coolest person in New York tech by Business Insider. “You can only provide so much instruction to a computer, so you need machine learning algorithms.”
As Shutterstock approaches its billionth image sold, it is using computer vision technology to extract out features of submitted images, generate better keywords, and translate descriptors into 21 different languages. To do that, their model, once operating off general themes, has had to grow into one capable of understanding concepts. At the same time, algorithms are now generating improved face recognition, automatic cropping, and recommendations for photographers seeking to fill gaps in to company’s vast catalog.
“We want to help businesses get better at using the right image at the right time,” Oringer said. “It’s all about matching the right content with the customer.”
As successful as Shutterstock has become, he recalled, it followed 10 failed startups—but he never gave up on finding a niche in the marketplace.
“If you’re thinking of starting a company, do it now, it’s only going to get harder later,” Oringer said. “Just try selling something, and don’t be afraid to fail.”