Samsung Patents a Self-Testing Video Decoder That Picks Its Own Fill Direction
Every time your TV or phone decodes a video, it has to guess how to fill in each tiny block of the image from the pixels around it. Samsung has filed a patent for a decoder that auditions every possible fill direction and picks the one that fits best, without needing a separate instruction from the encoder.
What Samsung's auto-direction video trick actually does
Imagine you're putting together a jigsaw puzzle, but instead of seeing the box art, you have to figure out the shape of each missing piece by looking at the pieces already placed around it. That's roughly what a video decoder does dozens of times per frame.
When your phone or TV decodes a video, it breaks each frame into tiny blocks and reconstructs them one by one. For each block, it needs to know which direction the surrounding pixels flow (horizontal stripes, diagonal lines, and so on) so it can make a plausible guess about what belongs inside. Normally, that direction is sent along with the video file itself, which costs extra data.
Samsung's new approach lets the decoder figure out the best direction on its own. It uses two strips of already-decoded pixels: one to generate predictions, and the other to check how good those predictions are. The direction with the lowest error score wins, and that's what gets used to reconstruct the block. No extra instruction needed.
… determining a directional mode for the current block, from among the plurality of directional modes, based on cost values corresponding to comparison results between each prediction sample set of the plurality of prediction sample sets and the other one of the first sample set and the second sample set …
Translation: The decoder tests different fill directions to see which one matches the surrounding pixels best.
How the decoder tests and scores each fill direction
The patent describes a decoding method built around intra prediction (the process of reconstructing a video block using only neighboring pixels already decoded in the same frame, rather than referencing other frames).
The decoder identifies two sets of neighboring pixels around the current block:
- First sample set: one strip of already-decoded pixels, used as the reference to generate candidate predictions.
- Second sample set: a separate strip of already-decoded pixels, used as the ground truth to score those candidates.
For each of several directional modes (think: horizontal, vertical, diagonal at various angles), the decoder runs intra prediction using the first sample set and generates a predicted version of the second sample set. It then compares each predicted version against the actual second sample set and assigns a cost value (a numerical error score, where lower means better). The direction with the lowest cost wins.
Once the best direction is selected, the decoder applies full intra prediction to reconstruct the actual current block. The key insight is that this approach uses already-available decoded data to make the direction decision locally, which means the encoder does not need to transmit the mode explicitly, saving bitrate.
… performing intra prediction by using one of the first sample set or the second sample set as a reference sample set according to each directional mode of a plurality of directional modes to obtain a plurality of prediction sample sets corresponding to the other one of the first sample set or the second sample set …
Translation: It generates multiple test predictions by using neighboring pixels as a reference guide for different angles.
What this means for video quality on Samsung devices
If this approach holds up in practice, it could let video encoders send slightly smaller files without sacrificing image quality, because the direction information does not need to be packaged and transmitted alongside the video data. For you, that could translate to sharper video at the same bandwidth, or the same quality at a lower data cost, which matters most on mobile connections where every kilobit counts.
the pattern in Samsung's video-codec filings points toward investing heavily in decoder-side intelligence, shifting work away from the encoder and the file itself. Whether this specific technique makes it into a shipping codec standard like VVC or a future successor depends on how well it performs in formal testing, but the direction is clear.
Samsung's 123rd filing we've tracked since May adds to earlier work like faster duplicate region spotting and two-stage image storage, all part of our camera sensor push coverage.
You would never notice this working. The improvement, if it ships, arrives invisibly: a video stream that holds its picture quality a little better when your connection gets spotty, or uses slightly less data to look the same.
Codec efficiency is where real-world video quality actually comes from. Dozens of small tricks like this one, accumulated, separate a TV that looks sharp at modest internet speeds from one that needs a much faster connection to reach the same result. Samsung makes both the chips and the TVs, so a niche technical filing like this has a plausible path into hardware you can actually buy.
This is a narrow, incremental improvement in a crowded field. If it gets folded into a future video standard, streaming on your next Samsung TV becomes imperceptibly better. That is a real payoff, even if you never know it happened.
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
49 drawing sheets from US 2026/0281330 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →