April 6, 2023 | 16948 Views | By Dan Harlacher
The human face is one of the most identifiable patterns in the world. As infants, it’s the first pattern we learn to recognize and remains the key to communication and social interaction throughout our lives. We can even discern facial landmarks in natural objects that do not resemble faces. Our ability to perceive the nuances of facial features is unparalleled, and we are highly attuned to even minor details, even in low-quality photographs. We might even be willing to overlook other imperfections in an image if the facial features are realistic. However, if the face is not convincing, it can ultimately detract from the overall quality of the photo, regardless of how well other elements render.
This is something we currently battle when resizing or upscaling photos. Broad-based super-resolution (upscaling) AI models, like the ones in ON1 Resize AI, are trained on millions of tiles of photos that represent all of the details and textures we find in everyday life. That’s how it can recognize small, compressed trees and reproduce them as bigger, sharper, and cleaner trees, for example. But when it comes to faces, you need a different kind of AI to reproduce them faithfully.
We have been developing a new AI model, coming very soon in a future free update for ON1 Resize AI 2023 and Photo RAW 2023, especially for enhancing faces. It can detect small faces that lack the native detail to be scaled traditionally, and then generate a new high-quality face that matches the original subject’s morphology, hue, and structure from the underlying photo. It’s nothing short of amazing! But seeing is believing, so look at these samples of our progress. Stay tuned to our updates by signing up for our newsletter as well!
Keep in mind these are preliminary results. We are still continuing to train the AI models that make the magic happen. You will see this technology in the field first in a future release of ON1 Resize AI, but I can imagine it being useful in other places as well.