Sony Patent Turns Event Pixel Noise Into Usable Luminance Data
Most camera engineers spend their careers fighting sensor noise. Sony just filed a patent that treats that noise as useful data, specifically, as a way to measure how bright a scene actually is.
How Sony's sensor reads light from its own noise
Imagine a camera that only wakes up when something in the frame changes, a ball moves, a light flickers, a person walks by. That's called an event camera, and it's far more efficient than a traditional camera that captures full frames 30 times a second. The catch: because it only tracks changes, it has no direct way to tell you how bright or dark the scene is at any given moment.
Sony's patent flips the usual script. Every pixel in this kind of sensor produces tiny false-trigger signals called noise events, and it turns out those false triggers happen more often when the pixel is exposed to more light. Instead of filtering that noise out, Sony's design counts it, then uses the count as a brightness reading for each pixel.
The result is a sensor that can track fast motion and reconstruct a full picture of scene brightness, without needing a separate traditional image sensor alongside it.
How the noise-event count maps to luminance per pixel
The patent describes a photodetection device built around event-based pixels, the kind used in neuromorphic or dynamic vision sensors. These pixels fire a signal (an "event") only when the light hitting them changes by a set threshold, up or down. That makes them very fast and power-efficient, but it also means they produce no output when a scene is static, leaving brightness information unrecorded.
The key insight here is that noise events (spurious, random pixel firings that have nothing to do with real scene changes) occur at a rate that tracks ambient illuminance. More light hitting a pixel means more thermal and shot noise, which means more false events per second. This relationship is predictable enough to be useful.
The device includes a luminance information generation unit that monitors each pixel's noise-event count over a time window and converts that count into a per-pixel brightness estimate. In other words:
- Real scene-change events carry motion and edge data
- Noise events, tallied separately, carry brightness data
- Together they give you a more complete picture than either alone
The patent covers the detection architecture and the logic for separating and processing these two event streams at the pixel level.
What this means for event-camera imaging in low light
Event cameras are already used in robotics, autonomous vehicles, and high-speed industrial vision because they handle fast motion and extreme lighting conditions far better than conventional sensors. Their one persistent weakness is the lack of absolute brightness information, which limits scene reconstruction and makes it harder to fuse event data with standard video. Sony's approach could close that gap without adding a second sensor or separate exposure cycle.
For Sony, which manufactures image sensors used in everything from smartphones to cars to medical devices, this is a meaningful extension of its event-camera work. If the technique can be made reliable across a wide brightness range, it could reduce hardware complexity and cost in systems that currently pair an event sensor with a traditional frame-based sensor to get both motion and luminance data.
This is a genuinely clever piece of sensor engineering. Using noise as a signal rather than an artifact to be eliminated is the kind of insight that looks obvious in hindsight but takes real expertise to formalize into a working architecture. Whether Sony can make the luminance estimates accurate enough for practical imaging tasks is the real question, but the underlying idea is solid and the problem it solves is real.
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 Sony patent news breaks.
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
24 drawing sheets from US 2026/0222702 A1 · click any drawing to enlarge
Show the other 16 sheets
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →