The dominance of Adobe Photoshop in the world of image editing may soon start to crack. A team of researchers from MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) has proposed an AI-assisted image editing tool which can automatically segregate each set of layers of an input image.
The MIT researchers working on the project have still not given a commercial name to the revolutionary image editing tool. For now, they’re chosen to call it by the foundational technology involved, Semantic Soft Segmentation (SSS).
Image Editing Tools
There are two main processes in editing an image – Selection and Compositing. Creating a precise selection is a tedious task, though Adobe Photoshop’s tools namely Magic Wand and Magnetic Lasso and Matting tools have existed for a long time to assist photo editors and ease out this process.
Even then delivering high-quality edited images is laborious and can only be done by skilled visual artists. At times, it becomes difficult for image editors to decide whether a pixel belongs to the background or the person’s hair.
“Instead of needing an expert editor to spend several minutes tweaking an image frame-by-frame and pixel-by-pixel, we’d like to make the process simpler and faster so that image-editing can be more accessible to casual users,” said YagizAksoy, an external researcher at MIT’s CSAIL.
The other team members of this project are Tae-Hyun OH and Wojciech Matusik of MIT CSAIL, Sylvain Paris from Adobe Research, and Marc Pollefeys from ETH Zürich and Microsoft.
How AI Image Editing Tool Works?
For SSS the team has developed a method to generate soft segments. In detail, these segments are “a set of layers that represent the semantically meaningful regions as well as the soft transitions [such as hair or grass] between them,” the researchers explained in the paper titled on the SSS technology itself and presented it at the annual SIGGRAPH Computer Graphics Conference in Vancouver the previous week.
In the AI-based image editing tool, the neural network automatically fuses“high-level and low-level image features in a single graph structure”. These segments are then segregated by solid colors including Red, Green, Blue, and Yellow.
The image editors can eventually use each colored segment for the purpose of editing or compositing altogether a new image. For instance, they can change the background or foreground in a picture or easily merge multiple layers from different images to create a new scene.
Here the artificial intelligence input completely eliminates the manual task of accurate selection in an image and therefore reducing the time during editing.
The MIT researcher team presently wants to employ this AI assisted image editing tool on image-sharing online platforms like Instagram and Snapchat, reported Techworm. They aim to make filters look more realistic, wherein the netizens can easily change the background of a picture in a selfie.
The next step for the researchers is to enhance the AI image editing tool’s capabilities with respect to matching colors, and elements like illumination and shadows.