Samsung Patents AI That Stops Blurry Doubles From Appearing in Your Photos

By Patentlyze Team · Updated Jul 21, 2026

When your phone stitches together multiple shots to brighten a dark photo, anything that moved between frames turns into a blurry ghost. Samsung's new patent describes an AI approach to detect and erase those ghosts before they ruin the final image.

What Samsung's multi-frame ghost-removal actually does

Imagine you're taking a photo in a dim restaurant. Your phone snaps several frames in a row and merges them to get a brighter, cleaner shot. But if someone walks through the frame between snaps, that person might appear twice, or as a blurry smear, in the final photo. That artifact is called a "ghost," and it's one of the trickier problems in smartphone photography.

Samsung's patent describes a system that looks at each extra frame, compares it to one "reference" frame, and builds a map showing exactly where things moved. The system then pre-blends each extra frame with the reference frame using that map, essentially softening the mismatched parts before the AI does its final merge. The result, in theory, is a single sharp image without the telltale double-exposure smears.

The AI model behind this was trained using synthetic motion maps, meaning engineers fed it thousands of artificially generated motion scenarios so it could learn to handle real-world movement without needing a massive library of real ghosted photos.

How the motion maps and blending network work together

The patent covers a processing pipeline for multi-frame photography, the technique where a camera captures several rapid-fire shots and combines them into one better image. The core problem it addresses is "ghosting," the visual artifact that appears when a subject moves between frames.

Here's the sequence the system follows:

• Capture multiple frames, designating one as the reference frame (the anchor image).
• Generate a motion map for each other frame by comparing it to the reference frame. A motion map is essentially a grid that marks which pixels moved and by how much.
• Use each motion map to pre-blend its corresponding frame with the reference frame. Areas flagged as high-motion get blended more heavily toward the reference, reducing the ghosting risk before the final merge.
• Feed the reference frame plus all these pre-blended frames into a blending network (an AI model) that produces one final image.

The training angle is notable. Rather than relying solely on real-world ghosted photos, Samsung trained the AI using synthetic motion maps, meaning artificially constructed movement data. This gives engineers precise control over what the model learns and avoids the difficulty of collecting labeled real-world examples of every possible ghosting scenario.

What this means for night mode and burst photography

Multi-frame merging is the engine behind most of what makes modern phone cameras impressive: night mode, HDR, low-noise shots. Ghosting has always been the soft underbelly of that approach. A better deghosting system could mean cleaner night-mode photos in busy or unpredictable scenes, like streets, concerts, or crowded rooms, without forcing you to hold your phone extra still.

For Samsung specifically, this fits into the ongoing camera arms race between Galaxy and iPhone flagships. Both companies keep pushing multi-frame computational photography further, and fixing ghosting is one of the remaining visible failure modes that regular users notice and complain about. If this technique makes it into a shipping camera app, you'd likely see the improvement most in low-light shots where people or cars are moving through your frame.

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