Samsung · AI / ML · Filed May 7, 2025 · USPTO published Nov 12, 2026 · This week's release

Samsung Patents AI That Rebuilds the Colors in Dark Phone Photos

Samsung's patent application describes using the same family of AI behind image generators to repair the colors in dark photos, one small patch at a time.

By Patentlyze Team · Updated Nov 12, 2026 · 3 min read

A side-by-side comparison reveals how Samsung's image processing restores lost colors and shadows in a dark photograph. Drawing from patent filing US 2026/0359932 A1. — A side-by-side comparison reveals how Samsung's image processing restores lost colors and shadows in a dark photograph.All 10 drawings ↓

Publication number: US 2026/0359932 A1(US20260359932A1)
Applicant: Samsung Electronics Co., Ltd.
Filed · Published: May 7, 2025 · Nov 12, 2026
Inventors: Devendra Kumar Jangid, Pavan Chennagiri Madhusudanarao, John Seokjun Lee, Hamid Rahim Sheikh
Status when we published: Waiting for an examiner
Document: 20 claims
US classification: 382/156
Examiner: WAMBST, DAVID ALEXANDER (Art Unit 2663)

AI image & video

What Samsung's dark-photo color fix actually does

Every time you snap a photo in a dim restaurant, your phone is guessing. Its sensor records only one color at each tiny dot (red, green or blue), and software fills in the other two by looking at the neighbors. In the dark, that guesswork gets sloppy, and a plain blue shirt can come out in patches of different blues.

Samsung's patent application swaps in an image-generating AI, the same family of technology behind AI art tools, to do the guessing. The AI works on a squeezed-down version of the photo, cleans it up step by step, then expands it back into a full-size image.

The filing also describes chopping the photo into pieces. The AI works on about half the picture, then the other half, then the whole thing, and the pieces get stitched together so the result looks consistent.

From the filing · CLAIM 1

… generating, by the at least one processing device, high-quality latent features based on the low-light latent features using a stable diffusion model and a conditional network; …

Translation: AI upgrades dark and grainy photo details into a clear picture using advanced diffusion technology.

How the AI cleans a photo piece by piece

The process starts with a raw low-light frame from the camera sensor. Sensors sit behind a pattern of red, green and blue filters (a Bayer pattern), so each pixel records only one color. The first step runs the frame through Bayer filtering to produce a linear sensor output, meaning values that rise in direct proportion to the light hitting the sensor. Several frames can optionally be aligned and blended first.

Next, an encoder (a neural network that compresses the image) shrinks the data into latent features, a compact summary of edges, textures and shapes. It uses convolutional layers, which scan small regions for patterns, and max pooling, which keeps the strongest signal in each small region.

A stable diffusion model (an AI that removes noise step by step to produce a cleaner result) then refines those features. A conditional network feeds it guidance from the original dark image so the output stays faithful to the real scene, and random noise can be added as part of the process. A decoder expands the cleaned features back into a full linear image, which can then be adjusted for display.

The filing also describes patch splitting: patches from about half the image, then the other half, then the whole image are each cleaned up and merged. The text says this improves consistency between patches. The same approach could also handle sharpening, denoising and de-blurring.

What this could mean for your night photos

For you, the payoff would be dim-scene photos with truer colors, such as a solid-color jacket staying one shade instead of splitting into blotches. The filing lists smartphones, tablets and cameras as possible devices, so the target is the everyday cameras people already carry. Dim scenes are a common weak spot for small camera sensors, so a fix there is easy to notice.

The evidence in the filing is a single example comparison against standard color reconstruction, with no timing or battery figures. It also says some of the work could run on a server instead of the device, which suggests heavy AI processing is a real question here.

Samsung's 169th filing we've tracked since May joins earlier work like one restoring blurry photo detail and one pausing cleanup for low battery in our camera sensor push.

Editorial take

If you shoot in dim bars, at concerts or on evening streets, you've seen this flaw: a jacket that should be one color comes out patchy and smeared. That is the failure this filing goes after, and you would spot the improvement right away when you zoom in on a photo.

The only proof in the document is one before-and-after example of clothing. There are no measurements, no speed figures and no word on whether a phone could run this on its own. Image-generating AI is heavy work, so the practical question is whether it fits in a device you carry in your pocket.

If it works, the benefit arrives with no extra effort from you: cleaner shadows and truer colors in the photos you already take. Treat it as a promising direction for night photos, with a lot left to prove.

Source: US 2026/0359932 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.

Get our take in your Top Stories

Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Samsung patent news breaks.

Add Patentlyze on Google One click.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.