Blog

Removing backgrounds with the help of computer vision

How Kuvatilaus.fi and Silo AI use the latest methods in computer vision for flawless image matting

Imagine you have a studio photo that you would want to blend on a different background. You open your photo editing tool, and use, depending on your skills, countless minutes or hours to individually do this to all your photos.

Now imagine you do this with help of the latest deep learning methods in computer vision that analyze every pixel so that not a single hair is missed. In the fully customized solution built in collaboration by Kuvatilaus.fi and Silo AI, you simply upload your photos, and in a matter of seconds, the AI technologies take care of meticulously removing the background.

Leveraging the latest deep learning methods in computer vision

Tuuli Mankki, Service Manager at Kuvatilaus.fi, wanted to offer their clients an AI solution with unmatched quality to help them with a daily and time consuming task of removing the background:

“We wanted professional photographers to have access to a high quality matting tool that would work flawlessly even with challenging details and high resolution images, such as 4K photographs. In Silo AI, we found an experienced partner with which it was possible to create a tailored AI solution that perfectly fits our needs and exceeds all our expectations.”, Mankki comments.

Deep learning to do image matting of 4K images in a matter of seconds

The first step in background removal is to use AI to create a trimap to separate foreground from background. Then, computer vision technologies are used to select the foreground and eventually blend it on the desired new background.

Image of a trimap prediction
A trimap shows the foreground in white, uncertain areas in grey and background in black.

4K images, pixel perfection and fast performance – around 10 seconds – posed a challenge for Silo AI’s AI Scientist Mikko Tukiainen and his team. To solve it, the team needed to leverage several tricks in data processing and a clever solution design to optimize the neural net.

“It was great to collaborate with Kuvatilaus.fi on a challenging project that required a lot of customization. Luckily, they gave us over 50 thousand example studio quality portraits to train the AI solution, and with sophisticated techniques such as data augmentation, we were able to make the AI solution robust, fast and working with top-notch quality.” Tukiainen comments.

Tukiainen compared the AI solution with 100+ test images with other background removal solutions out there.

“In order to match the desired precision level, we needed to process and augment artificial images used in training with features like vignette or tilt shift blur as well as noise augmentation. As a result the matting tool provided by Kuvatilaus.fi now has the best performance on the market.”

Image shows progress of background removal before noise augmentation
Image shows progress of background removal after noise augmentation
Quality before vs. after noise augmentation. Mikko and his team of computer vision experts used techniques such as noise augmentation to improve the AI solution’s performance.

You may read more about Kuvatilaus.fi’s matting tool here & check out their video on the new feature here (both in Finnish).

---

Image of Pertti Hannelin

Would you like to discuss leveraging computer vision for your business? Get in touch with our VP, Business Development, Pertti Hannelin on LinkedIn or via email.

About

No items found.

Want to discuss how Silo AI could help your organization?

Get in touch with our AI experts.
Author
Authors
Pauliina Alanen
Former Head of Brand
Silo AI

Share on Social
Subscribe to our newsletter

Join the 5000+ subscribers who read the Silo AI monthly newsletter to be among the first to hear about the latest insights, articles, podcast episodes, webinars, and more.

What to read next

Ready to level up your AI capabilities?

Succeeding in AI requires a commitment to long-term product development. Let’s start today.