Samsung Patents a Video Compression Method That Learns From Already-Decoded Frames
Every time your phone encodes a video, it has to make thousands of guesses about what the next block of pixels will look like. Samsung's new patent proposes a way to make those guesses by reading clues already baked into the blocks it just finished.
How Samsung's gradient-guided video prediction works
Imagine you're filling in a crossword puzzle. Once you've filled in a few answers, you use those letters to help guess the next word instead of starting from scratch. Video compression works similarly: the encoder splits an image into small blocks and tries to predict each block's contents so it only has to store the difference between the prediction and reality.
Samsung's patent describes a system where, before predicting a block, the decoder looks at the direction of edges and patterns in the blocks it already handled. From those, it figures out which prediction direction (say, diagonal or horizontal) is most likely to work well for the current block, and how strongly that direction dominates.
The practical payoff is that the decoder spends less effort on wrong guesses, which can mean better image quality at the same file size, or the same quality at a smaller file. That matters most for things like video calls, streaming, and on-device recording.
How previous blocks steer the intra prediction mode choice
The patent describes a video decoding method centered on intra prediction (predicting a block's pixel values purely from other blocks in the same frame, rather than from other frames in the video).
The core steps are:
- Collect gradient information (measurements of how pixel intensity changes direction, essentially edge-direction statistics) from several previously decoded blocks in the same image.
- Filter that information to only keep values that meet a predetermined condition, such as being strong enough or consistent enough to be useful.
- Combine those filtered values into a gradient summary for the block currently being decoded.
- Use that summary to select an intra prediction mode (one of many standardized directions the decoder can use to extrapolate pixel values) and generate a prediction block.
- Compare the prediction to the actual encoded data and reconstruct the final block from the difference.
The two key numbers the system tracks are a first value (which prediction direction the neighboring blocks favored) and a second value (how strongly or consistently they favored it). Together, these steer the mode choice for the current block.
What this means for video quality and streaming efficiency
Video compression is one of the most compute-intensive tasks that runs on consumer hardware, from phone cameras to streaming servers. Any improvement in how accurately a codec predicts blocks translates directly into either smaller file sizes or sharper images at the same bitrate. Samsung is a major supplier of both the chips that run video codecs and the cameras that capture the footage, so patent filings in this area often point toward future codec tuning in their chipsets or camera software.
For you as a viewer or content creator, the downstream effect could be smoother video on slow connections or less storage eaten up by recorded footage. That said, codec patents are incremental by nature, and this one describes one piece of a much larger puzzle.
This is a narrow, incremental codec patent in a field Samsung has been filing in for decades. It is not unimportant: video compression underpins enormous amounts of everyday technology, and marginal gains compound at scale. But readers should not expect a visible product moment from a single filing like this.
The drawings
41 drawing sheets from US 2026/0214215 A1 · click any drawing to enlarge
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.
Editorial commentary on a publicly published patent application. Not legal advice.