Google Patents a Video Filter That Uses Color Data to Fix Compression Blur
Every time you stream a video, compression throws away information to keep file sizes manageable. Google has filed a patent for a filter that uses color data the codec already has to recover some of what was lost.
What Google's cross-color video filter actually does
A security camera records a crowded intersection all day. By the time that footage reaches your screen, a compression algorithm has squeezed it down to a fraction of its original size, and the result can look blurry or blocky. That tradeoff happens with every video you watch, from a YouTube clip to a video call.
Google's new patent describes a way to clean up that blurriness after a video frame has been decoded. Instead of only looking at the blurry pixel and trying to guess what it should have been, the filter also looks at a different color channel in the same frame. Most video formats store brightness and color in separate layers, and those layers often carry overlapping detail. The filter uses that overlap to make a better guess about the correct pixel value.
The result, in theory, is a sharper reconstructed frame using information the decoder already has, with no need to transmit extra data.
generating a reconstructed frame by decoding an encoded frame from a compressed bitstream, wherein the reconstructed frame comprises a first color plane and a second color plane; and obtaining an enhanced reconstructed frame by applying a filter to at least one pixel of the first color plane of the reconstructed frame using at least one pixel value of the second color plane.
Translation: Decoding compressed video creates frames with multiple color planes, and applying a filter across these planes cleans up the image.
How the filter borrows pixel data across color channels
Modern video codecs split a frame into at least two color planes: a luma plane (brightness, typically called Y) and one or more chroma planes (color hue and saturation, typically called Cb and Cr). When a video is compressed and then decoded, each plane is reconstructed separately, and compression artifacts can appear in any of them.
This patent describes a Wiener filter (a classical signal-processing tool that estimates a clean signal from a noisy one by weighting nearby samples) applied to pixels in one color plane while also feeding in pixel values from another color plane. In plain terms: when the filter is trying to decide what the true brightness of a pixel should be, it is allowed to peek at the color information sitting in the same location of a different channel.
The filter coefficients (the weights that determine how much each neighboring pixel contributes) are designed or transmitted as part of the video stream, so both the encoder and decoder agree on how the filtering should work.
Key elements include:
- Decoding a compressed frame to produce a reconstructed frame with separate color planes
- Applying a filter to at least one pixel in the first color plane
- Using pixel values from the second color plane as additional inputs to that filter
- Outputting an enhanced reconstructed frame with reduced compression artifacts
What sharper compressed video means for everyday streaming
For the average viewer, this kind of filter is the difference between a video call that looks like a watercolor painting and one where you can actually read the text on someone's whiteboard behind them. Compression artifacts are most visible in high-motion scenes or at lower bitrates, exactly the conditions you hit on a slow connection or in a crowded Wi-Fi environment.
Google controls both the VP9 and AV1 open video codecs, which power YouTube and a large share of web video. Google's steady work on in-loop and post-processing filters suggests this kind of enhancement is being built into the codec pipeline itself, not bolted on as an afterthought. If this filter lands in a future codec version, the improvement would be automatic for every viewer without any change on their end.
This is the 689th Google filing in our Google coverage since May, adding to a body of work that includes a video motion patent and a skin-tone sensor fix.
If you have ever watched a video go blotchy or soft during a slow connection moment, this addresses exactly that. Google's approach uses color information the device already decoded to clean up the image, rather than just accepting the blur as an artifact of compression.
The catch is that this cleanup happens on the viewer's device or server while the video plays, and that extra work costs battery and processing power. On a phone handling a sharp, high-resolution stream, those costs are real.
When it works well, you would never notice it, which is the point. Videos would simply hold together better in the moments compression handles worst, and that quiet improvement is what most people actually care about.
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.
The drawings
8 drawing sheets from US 2026/0270419 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →