Samsung Patents a Sampling-Grid Method for Pixel-Level Image Reconstruction

By Patentlyze Team · Updated May 22, 2026

Samsung is patenting a way to rebuild images pixel-by-pixel using a grid that already knows where each reconstructed pixel should land — a low-level but potentially powerful approach to image upscaling, correction, or super-resolution.

What Samsung's offset sampling grid actually does to your photos

Imagine you took a photo and some pixels came out blurry, misaligned, or just wrong. Instead of guessing how to fix them on the fly, what if your phone had a pre-built map telling it exactly where every pixel should be — and then just moved them there? That's the core idea here.

Samsung's patent describes a system that takes an input image and runs it through a sampling grid — essentially a coordinate map layered on top of a standard pixel grid. Each position in that map carries "offset information," which tells the system the precise location each pixel should be reconstructed at. The result is a reconstructed image where pixels have been repositioned or regenerated based on that pre-computed spatial guidance.

This kind of approach shows up in image upscaling, lens-distortion correction, and neural super-resolution pipelines. It's a building block, not a finished feature — but building blocks are what camera systems are made of.

How the pixel grid and offset data rebuild each image

The patent describes a three-step image processing pipeline:

• Input image acquisition — the system takes in a source image, which could come from a camera sensor, a video frame, or any pixel buffer.
• Sampling grid construction — a pixel grid (a regular, predefined coordinate lattice) is augmented with offset information, meaning each grid point is tagged with a displacement vector or target location that says "this pixel belongs over here, not where the grid originally placed it."
• Grid application and reconstruction — the offset-augmented sampling grid is applied to the input image, pulling or interpolating pixel values from the shifted locations to produce a reconstructed image.

The key technical idea is that the offset information encodes reconstruction location information per pixel — meaning the system doesn't just apply a uniform warp or scale; it can have a unique spatial correction for every single grid point. This is the same conceptual backbone used in deformable convolutions (a neural-network technique where filters learn to sample irregular spatial locations rather than fixed rectangular ones) and in spatial transformer networks.

The claim is intentionally broad — it covers the method regardless of how the offset grid is generated, which leaves room for both learned (neural) and hand-crafted (geometric) offset sources.

What this means for Samsung's camera and display pipeline

For Samsung, which ships hundreds of millions of camera-equipped devices per year, even small improvements in pixel reconstruction fidelity can meaningfully affect perceived image quality. This kind of spatially-aware grid approach is a core primitive in super-resolution, lens-distortion correction, and image stabilization — all areas where Samsung competes aggressively with Apple and Google.

The broad claim language is worth noting: by patenting the general method of applying an offset-assigned sampling grid to reconstruct pixels, Samsung is staking ground across a wide range of image processing use cases. Whether this ends up in a dedicated ISP (image signal processor) chip, a neural super-resolution model running on-device, or a display upscaling pipeline, the foundational mechanic is the same.

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