Nvidia · Filed Feb 23, 2026 · Published Jul 2, 2026 · verified — real USPTO data

Nvidia Patents a Multi-Pass System for Cleaning Up Color Artifacts in Streamed Images

When video is compressed and streamed, some pixels end up the wrong color. Nvidia's new patent describes a layered process that finds those bad pixels and replaces them with values that look right, without accidentally smoothing over real edges in the image.

Nvidia Patent: Removing Streaming Image Artifacts — figure from US 2026/0187971 A1
FIG. 1A — rendered from the official USPTO publication PDF.
Publication number US 2026/0187971 A1
Applicant NVIDIA Corporation
Filing date Feb 23, 2026
Publication date Jul 2, 2026
Inventors Andrew Russell, Prabindh Sundareson
CPC classification 382/167
Grant likelihood Medium
Examiner CENTRAL, DOCKET (Art Unit OPAP)
Status Docketed New Case - Ready for Examination (Mar 25, 2026)
Parent application is a Continuation of 18151653 (filed 2023-01-09)
Document 1 claims

What Nvidia's artifact-removal pipeline actually does

Imagine watching a show on a slow internet connection and noticing a patch of the image looks slightly off, maybe a gradient that has visible banding, or a patch of skin tone that looks like a faint grid pattern. That kind of glitch is called an artifact, and it shows up a lot when images are compressed and sent over a network.

Nvidia's patent describes a system that automatically hunts for those bad patches in a streamed image. It runs the image through several analysis steps to figure out which areas have real color detail (like the sharp edge of a door frame) and which areas are just compression noise. It treats those two types of areas differently so it doesn't accidentally blur real detail while trying to fix the noise.

Once it knows where the problem pixels are, it works backward through its own analysis to calculate better color values for each one. It can then use a technique called dithering, which spreads tiny color errors across neighboring pixels so your eye blends them into a smooth result instead of seeing a visible pattern.

How the forward and backward passes fix bad pixels

The system runs the incoming image through a series of forward passes, each one generating a simplified version of the image at a different level of detail. By comparing these levels, it can distinguish between two things that look similar in a compressed image: a true color edge (where two genuinely different colors meet) and an artifact (where compression has introduced a false color pattern).

Once those areas are labeled, the system runs a series of backward passes, working from the most simplified level back to the original. At each step, it recalculates the color values for artifact-affected pixels using the color information from the level above. This layered correction means the fix is informed by both local pixel data and broader context from the image.

Finally, the corrected pixel values can go through dithering, a classic technique where tiny rounding errors in color are distributed across nearby pixels rather than concentrated in one spot. The result looks smoother to the human eye because vision naturally averages neighboring pixels together.

  • Forward passes: build a multi-level map of the image to locate artifacts
  • Backward passes: recalculate correct color values working from coarse to fine
  • Dithering: spreads residual color error so it becomes invisible

What this means for game streaming and remote graphics

This kind of pipeline is most useful in cloud gaming and remote desktop streaming, where images are compressed hard before being sent to your device. Artifacts are a persistent complaint in those services, and fixing them on the receiving end without blurring real image detail is genuinely difficult.

For Nvidia, which runs the GeForce NOW cloud gaming service and sells GPUs used in data-center streaming setups, having a patented artifact-removal method baked into its processing stack would give it a quality argument over competitors. If this lands in a driver or streaming codec update, you might notice cleaner gradients and skin tones in streamed games without any change to your internet speed.

Editorial take

This is a solid, specific patent that addresses a real problem anyone who has used GeForce NOW or a remote desktop has noticed. The multi-pass forward-backward architecture is a reasonable engineering approach, not just a vague idea. Whether it ships in a meaningful form depends on whether the performance cost fits inside the tight latency budgets of live game streaming.

Which company should we read for you?

We track 17 companies here. Pro is the same weekly breakdown for any company you choose, delivered privately. Type a name and we'll scope it and send you a quote.

Get one Big Tech patent every Sunday

Plain English, intelligent commentary, no hype. Free.

Source. Full patent text and figures from the official USPTO publication PDF.

Editorial commentary on a publicly published patent application. Not legal advice.