Shutterstock is one of the largest stock photography and video sites on the web. Finding the right content for your next project can be challenging, particularly when you need the right video clip (harder than images). Today, Shutterstock has announced Reverse image search for Video.
Leveraging machine learning the system uses a custom-built convolutional neural network the system compares the image you supply it, to the massive collection of content ( 280 million images and more than 15 million video clips) in their system and returns videos that match your image. Pretty neat right?
As long as you have a photo, frame, or screenshot that illustrates the clip you need, there’s a strong chance you’ll find the perfect clip thanks to the algorithm offering videos with a similar look and feel to the uploaded image.
If you create videos in your professional life, or for your YouTube channel, then you’ll love this new feature as it dramatically speeds up the task of finding the right content.
Shutterstock has had reverse image search since 2016 and heard great feedback from users, so the company then embarked on the journey of delivering the same feature for video.
The system breaks down each frame of a video at the pixel level into their principal features. It recognizes shapes, colours, locations, even down to camera framing.
I took it for a test drive and below are a few samples of the results I got from some images I had laying around. While some of the results had a couple of questionable results, the majority of results were on-point. This feature is a pretty big success and definitely something that would reduce your time to find the right clip.
You can read more about Shutterstock’s Machine Learning efforts here.