The application makes use of the recently introduced computer vision technology to visually locate similar images like the uploads, and mechanically generates relevant and previously successful keywords for the contributor to select from. The application uses its artificial intelligence technology to analyze whether metadata and pixel data are working together to create the most accurate keyword suggestions.
In an interview with VentureBeat Lawrence Lazare, Product Director for Shutterstock said:
Previously, Shutterstock has used this approach for the reverse image search it rolled out earlier this year, and deep learning is also behind the tool in the Shutterstock app that picks out visually similar images.
Shutterstock contributors can add upto 50 keywords to accompany any given image they upload to Shutterstock, which helps customers, discover the imagery best suited for their needs. Shutterstock claims to have over 100,000 contributors which may be advantageous because of these added features.
Jon Oringer, Founder and CEO of Shutterstock said:
We continue to invest in our Artificial Intelligence tools, which allow us to enhance the Shutterstock experience. In this instance, we are helping our artists and photographers focus on the job they enjoy most. Keyword suggestions and other innovative tools built on our custom computer vision technology are instrumental in bringing efficiency and accuracy to the contributor experience, helping them earn faster.
According to Shutterstock, through machine learning, the new keyword suggestion tool analyzes the metadata in Shutterstock’s collection of over 90 million images to generate relevant tags for the image uploads. For instance, a user has clicked a set of nature photos and wants to upload the entire collection to Shutterstock. With this new feature, the user has no longer have to guess which tags will lead customers towards the images. Instead, the keyword suggestion tool will recommend relevant keywords for the images and the user will be prompted to select the best fits.
“A few years ago, though, ‘all the big companies came’ to Shutterstock, because these companies had lots of data, specifically images, that can be useful to train on. Nowadays, Shutterstock has 93 million images to build up the understanding of its neural network for automatic keyword generation,” added Lazare.
