Lightroom AI boosts image quality 30 percent, Adobe says

Lightroom’s “enhance details” uses AI to improve sharpness and color in areas of fine detail, Adobe said.


Adobe

After training an AI system on a billion photos, Adobe has reworked a fundamental part of digital photography for a 30 percent increase in image quality in its Lightroom software.

The improvement comes in new Lightroom releases Tuesday with an option called “enhance details” that can break past ordinary image-quality limits in areas with lots of fine details. It’ll take significant computing horsepower and benefit from a fast graphics chip, though, Adobe said.

“Enhance details will deliver stunning results including higher resolution and more accurate rendering of edges and details, with fewer artifacts like false colors and moiré patterns,” said Josh Haftel, Adobe’s Lightroom product manager, in an explanation of the technology.

The development shows the continuing improvements to the higher-end raw photo format. Most of us are fine shooting JPEG images, or perhaps their space-saving alternative, HEIC. But raw images preserve the greatest flexibility, which means that raw photo you shot years ago can look better as software improves. Better raw technology also has upgraded older photos by removing more noise, the distracting color speckles that can mar photos shot in dim conditions.

The enhance details feature also shows how a technology called computational photography is helping to produce better photos from our camera hardware. Expect more of that in future cameras and photo-editing software.

Upgrade for Lightroom CC

Adobe has two versions of Lightroom for Windows and MacOS personal computers, the fuller-featured Lightroom Classic CC and the newer Lightroom CC designed to work better in conjunction with cloud computing and mobile devices. The new enhance details feature is in both versions.

Lightroom is Adobe’s software for editing and organizing photos.


Screenshot by Stephen Shankland/Techhnews

And Lightroom CC also gets an upgrade to stack up better against Lightroom Classic CC: the ability to merge multiple photos into a single panoramic or HDR image. High dynamic range photography is designed to capture a wider range of exposure data so photographers can have details in both murky shadows and bright highlights. Lightroom CC, also following Classic, now can combine the two approaches with HDR panoramas, too.

Lightroom CC also gets two other useful editing features from Classic: the targeted adjustment tool that lets you adjust attributes like color and exposure for a specific patch of a photo, and an indicator to show “clipping” — areas of the photograph that are completely washed-out white or blocked-up black. Clipping is an indication that you could be losing details you might want to preserve.

Better Bayer with AI

To understand why enhance details works, you have to understand a little about how digital cameras work. Each pixel in a digital photo records numeric values for the amount of red, green and blue, but the digital image sensor that actual records that data only captures a single color for each pixel — red, green or blue.

A checkerboard-like grid of red, green and blue colors called a Bayer pattern determines which color each pixel records. Then a process called demosaicing calculates the missing values so each pixel gets values for all three colors.

The Bayer pattern, and a cousin called X-Trans used in Fujifilm cameras, determine how each pixel site on a camera image sensor records only red, green or blue light. A camera or photo editing software must reconstruct red, green and blue data for every pixel, a process called demosaicing.

The Bayer pattern, and a cousin called X-Trans used in Fujifilm cameras, determine how each pixel site on a camera image sensor records only red, green or blue light. A camera or photo editing software must reconstruct red, green and blue data for every pixel, a process called demosaicing.


Adobe

That calculation is just a computer’s best guess at the missing color values, though. It’s not hard in areas of uniform color, like a blue sky, but where an image has to capture fine details like fabric or hair, demosaicing algorithms can introduce errors that show up as weird patterns, unnatural edges or wrong colors.

Enter artificial intelligence — a technology domain Adobe describes with its Sensi brand name.

The AI computing revolution or machine learning today uses a computing technology called neural networks that are modeled loosely on the human brain. You can train a neural network to recognize patterns like human faces or take actions like rejecting incorrectly manufactured parts. To do so requires processing large amounts of training data — photos labeled in advance with human faces, for example.

In Adobe’s case, it trained its Lightroom neural network on more than a billion photos that are a challenge for traditional demosaicing approaches. The result is a better guess at just how to reconstruct red, green and blue color data and therefore a sharper image.

Adobe’s enhance details technology works with the Bayer pattern most digital cameras use and a different pattern called X-Trans in many Fujifilm cameras. Adobe has faced persistent complaints that its demosaicing software doesn’t handle X-Trans sensors as well as rivals like Phase One’s Capture One Pro.

NASA turns 60: The space agency has taken humanity farther than anyone else, and it has plans to go further.

Taking It to Extremes: Mix insane situations — erupting volcanoes, nuclear meltdowns, 30-foot waves — with everyday tech. Here’s what happens.

Source link


Leave a Reply

Subscribe to our newsletter

Join our monthly newsletter and never miss out on new stories and promotions.
Techhnews will use the information you provide on this form to be in touch with you and to provide updates and marketing.

You can change your mind at any time by clicking the unsubscribe link in the footer of any email you receive from us, or by contacting us at newsletter@techhnews.com. We will treat your information with respect.

%d bloggers like this: