By Yaroslav Kuflinski
Guest Contributor
For millions of people worldwide, Adobe Photoshop is an infinitely useful tool that is used to edit images and bring out their best features. However, its use can be time-consuming, and it is very much a specialist piece of kit that takes a while to learn about and use proficiently. But this may be about to change.
MIT’s Computer Science and Engineering Lab have recently created Semantic Soft Segmentation (SSS)—an AI-assisted image editing tool that automates the selection and isolation of individual objects within images. This makes the task of photo editing much easier and more accessible—the project’s ambition is to develop it to the point where editing images takes nothing more than a simple click.
This opens vast opportunities for people with cameras and computers—both for those who want to create art and for those who would like to fake reality.
How SSS Works
In their paper, the team behind SSS described how it uses AI to separate individual objects within an image so that they can easily be manipulated. For example, by using SSS, you could isolate an individual from a group of people and then superimpose this individual onto a different background.
Image Source: MIT CSAIL
To do that, the system analyzes data about the colors and textures in an image and then compares it to the object labels present in this image. The latter are derived by a neural network—a machine learning model that can classify objects in images.
Whilst processing this information, the system detects soft transitions such as vegetation and fur or hair. This is usually the hardest part of image segmentation as the edges of an object are blurred into the background. SSS analyzes these transitions and effectively separates objects from one another and from the background creating semantic soft segments—layers that correspond to meaningful regions within an image.
What this means is that the end-users don’t need to change transitions themselves or make modifications to individual layers. This means that editing tasks such as color adjustments and changing backgrounds are made a lot easier because the input of AI eliminates the manual task of accurately selecting specific elements within an image.
What SSS Can Do
Researchers are currently working on making image computation using SSS much quicker and making it easier for the system to match colors and handle other elements such as shadows, illuminations, and shading. When the capabilities of SSS begin to improve and the technology is enhanced, the potential applications for it are limitless.
The easiest potential application for SSS in its current form exists within social media platforms such as Instagram and Snapchat where users can apply various filters to their pictures. For example, users could change the backgrounds on their Snapchat pictures or simulate different types of camera by using SSS.
A more creative way to use SSS may be related to visual arts and design. The ability to the change background in a realistic way provides for multiple interpretations of well-known ideas and compositions. And this leads us to the last and the most disturbing potential application of the new image editing technique—the ability to produce more credible fakes.
Fake news is already a major epidemic on digital media, duping millions of people worldwide with fabricated stories. Those range from relatively innocent fool-day jokes to complex manipulations designed to change the opinions of the audience on important topics such as international politics and wars or man-made pollution and natural disasters. The virtually limitless possibilities of the SSS technology and many its alikes, which undoubtedly will follow soon, may make people question the reality of the world we are living in—like in the Matrix movie or the Black Mirror show.
We have seen time and time before that when new and revolutionary technologies hit the market, there are many people out there who are quick to abuse and manipulate them for their own gain or satisfaction, and this is something which, should MIT’s tool or another iteration of it become mainstream, will prove to be a major problem, especially since it is backed by AI.
What’s Next for Semantic Soft Segmentation?
There is no denying that SSS is a major technological breakthrough that has the potential to be truly revolutionary and make the job of accurately editing images a breeze.
As this tool gets better over time, it is virtually guaranteed to be something that many designers and other creatives take an interest in; the new features seen makes image editing tasks which would have otherwise consumed a large proportion of time easier.
Although the technology and AI behind it are far from perfect, it is only going to improve as the AI learns from its continued use and the brains behind it work hard to make improvements. Funnily enough, the biggest challenge is not making the technology better but finding a way to prevent it from slipping into the wrong hands and curtailing the possibility of it being used to fuel fake factories.
Almost any technology advanced enough may be used both for good and for bad—it depends only on the people that use it. So there’s no reason to fear new opportunities provided by AI developers—embracing AI and devising innovative ways to use it you can make this world better and amend for its possible evil applications. It’s only up to you which side of the Force to choose.
Information about the author
