Samsung Patents a System That Maps Skin Tone Across Your Whole Body From Your Face

By Patentlyze Team · Updated May 22, 2026

Samsung has filed a patent for a technique that figures out a person's natural skin tone from their face alone — then uses that as a reference to consistently segment and classify every other skin-colored region on their body in a photo.

What Samsung's face-to-body skin mapping actually does

Imagine you take a photo at a pool party. Your phone's AI tries to select just the people in the shot — but it struggles to tell where skin ends and a similarly-toned towel begins, especially on arms, legs, and torsos far from the face.

Samsung's patent tackles this by treating the face as a reliable anchor. Because faces are usually well-lit and unobstructed, the system first classifies the skin tone there — combining a neural network's read on facial attributes (like texture, lighting, and structure) with a color component classification (the actual hue and tone values). That combined "fingerprint" is then pushed outward to identify matching skin regions on the rest of the body.

The result is a more consistent, person-specific segmentation mask — one that knows your skin tone, not just a generic "skin color" range. This kind of precision matters a lot for photo editing, background blur, and beauty filters that should adapt to individual people rather than average them out.

How the classifier combines color and attributes to propagate tone

The patent describes a two-track classification pipeline that runs on a detected person instance inside an image.

Track 1 — Attribute classification: A neural network analyzes the facial region to predict high-level attributes. Think of this as the system learning what kind of face it's looking at — skin texture, tone family, lighting conditions — and producing a soft probability score for a set of defined classes.

Track 2 — Color component classification: Separately, the system runs pixel-level color analysis on the same face crop. It uses k-means clustering (a technique that groups pixels into color buckets by similarity) in LAB color space — a perceptual color model designed to match how human eyes distinguish hues — to find the dominant skin colors. Confidence thresholding filters out noise and background bleed.

Those two outputs are then combined into a single classification for the facial region. That fused label — carrying both perceptual color data and learned attribute context — is then propagated to the rest of the body. Regions outside the face that match the established skin-tone fingerprint get assigned to the same person-specific class, building a full-body natural-skin segmentation map.

The claim language also references a "remaining natural portions" concept, meaning the system is specifically targeting skin and similarly natural body surfaces, not clothing or accessories.

What this means for Samsung's camera and photo-editing pipeline

For Samsung's camera and image-processing stack, consistent person segmentation is the foundation of a lot of downstream features — portrait mode, background replacement, skin-tone-aware exposure, and selective editing tools. A system that anchors body segmentation to a face-derived color fingerprint would make all of those features more accurate for people across the full range of human skin tones, rather than relying on a single global "skin" hue range that tends to work better for lighter tones.

There's also a fairness dimension here. Many segmentation systems historically perform worse on darker skin tones because their training data was skewed. A person-specific, face-anchored approach that learns your particular tone before classifying the rest of you is structurally better positioned to be equitable — though whether it actually closes that gap depends entirely on implementation and training data quality.

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