💻 computer science

Shared Damage-Field Control for Real-Time Appearance Modeling of Severely Damaged Hair

This paper presents a real-time, GPU-based procedural framework that uses a shared spatial damage field to drive spatially correlated geometric and optical changes for the interactive appearance modeling of severely damaged hair without requiring dense auxiliary geometry or high-resolution textures.

Original authors: Min-Song Park, Jong-Hyun Kim

Published 2026-11-12

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Min-Song Park, Jong-Hyun Kim

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the world of computer graphics, creating realistic hair has long been a pursuit of perfection. For decades, digital artists have focused on rendering healthy, idealized strands that catch the light with a glossy sheen and flow with smooth, predictable curves. This work relies on understanding how light bounces off thin fibers, how it passes through them, and how they scatter in groups. The goal has always been to make digital hair look alive and well-cared-for. However, real life is rarely perfect. Hair suffers from sun exposure, chemical treatments, and heat styling, leading to a specific set of visual problems: the strands become frizzy and disorganized, the ends turn a lighter, bleached color, the shiny highlights fade into a dull matte, and the tips split and break apart. Until now, simulating this kind of severe damage in real-time applications, such as video games or interactive digital humans, has been difficult. Artists often had to control each of these broken features separately, which made it hard to keep the damage looking consistent across the entire head of hair.

A team of researchers at Inha University has developed a new way to model this damaged appearance that solves the problem of consistency while remaining fast enough for interactive use. Instead of treating frizz, discoloration, dullness, and breakage as four separate tasks, they created a single, shared "damage field." Imagine this field as an invisible map painted over the hair that tells the computer exactly where the damage is located. This map is built from three simple pieces of information: how far down the strand you are from the root to the tip, whether that specific strand is on the outside of the hairstyle where it gets more exposure, and a bit of random variation to ensure no two strands look exactly alike. This shared map acts as a master controller. When the user decides to increase the level of damage, this single signal travels through the system and triggers four different responses simultaneously, but each response reacts at its own speed and intensity.

The system works by dividing the work between two stages of the computer's graphics processor. In the first stage, the computer looks at the shape of the hair and uses the damage map to gently push the strands outward, creating a frizzy volume that grows larger toward the tips. It does this in a way that keeps the strands flowing naturally, avoiding the jagged, torn look that simpler methods produce. In the second stage, the computer handles the surface appearance. It gradually lightens the color of the hair, starting at the tips and moving upward, mimicking the way sun and chemicals bleach hair over time. At the same time, it softens the sharp, bright reflections that healthy hair shows, replacing them with a duller, more diffuse glow. Finally, it addresses the broken ends. Instead of simply cutting the hair short, the system makes the very tips of the strands fade away and become transparent, creating the illusion of split ends and irregular tearing without actually needing to build complex new geometry.

The researchers tested this method on various hairstyles, including long straight hair, short layered cuts, and wavy styles. They found that the approach works seamlessly across all of them. As the damage level increases, the hair naturally transitions from healthy to severely damaged. The frizz appears early, the color lightens gradually, the shine fades quickly, and the broken tips only become obvious when the damage is severe. This progression feels natural because the system understands that damage usually starts at the ends of the hair and works its way up. The method is also highly efficient. On a powerful computer workstation, adding this full damage simulation to a standard hair rendering process only increased the time it took to draw each frame by about 12.6 percent. This small cost means the effect can be used in real-time applications without slowing them down.

It is important to note that this system is designed as an appearance model, not a scientific simulator of biology. The researchers are not trying to predict the chemical reactions that happen inside a hair strand or the exact biological process of how a cuticle breaks down. Instead, they are creating a visual tool that allows artists to control the look of damaged hair with a single, intuitive dial. The system does not require artists to paint detailed texture maps or build extra 3D models for every strand. By using a shared control signal to coordinate geometry, color, light, and transparency, the method ensures that all the signs of damage stay spatially connected. If the hair is frizzy in one spot, it is also likely to be discolored and dull in that same spot. This creates a cohesive and believable result that would be very difficult to achieve by adjusting each feature independently.

The study confirms that separating the location of damage from the specific visual response is a powerful way to handle complex appearance changes. The researchers demonstrated that this approach scales well, meaning it works just as effectively on a simple hairstyle with a few thousand strands as it does on a complex one with hundreds of thousands. While the current results are based on computer-generated hair assets, the framework is intended to be a flexible tool for digital artists. It offers a way to visualize the effects of hair care or neglect in interactive environments, from video games to virtual production, without sacrificing performance or visual quality. The work suggests that by sharing a common control signal, complex visual degradation can be managed with simplicity and precision, bringing digital hair one step closer to the imperfect reality of the human experience.

Technical Summary: Shared Damage-Field Control for Real-Time Appearance Modeling of Severely Damaged Hair

Problem Statement

Real-time hair rendering has historically focused on healthy or idealized fibers, emphasizing anisotropic reflection, transmission, and multiple scattering. However, severely damaged hair presents a distinct visual challenge characterized by the simultaneous occurrence of geometric disorder (frizz), tip-biased discoloration, reduced gloss (highlight degradation), and strand-end fragmentation (split ends).

Current approaches often treat these cues independently. Controlling them separately creates a significant authoring burden to maintain spatial consistency between geometry and shading. Furthermore, simulating the underlying chemical or biological processes of hair damage is computationally prohibitive for real-time applications and unnecessary for visual fidelity. The paper addresses the need for a compact, real-time procedural model that coherently controls these visible consequences on existing strand or hair-card assets without requiring dense auxiliary geometry or high-resolution damage textures.

Methodology

The proposed framework utilizes a Shared Damage-Field Control system. The core concept is a spatial damage field, $D_{local} \in [0,1]$ , defined over the hair geometry. This field determines where damage is expressed, while component-specific response functions determine how strongly geometry, color, reflectance, and opacity react at each location.

1. Spatial Damage Field Formulation

The local damage field is constructed from three primary inputs:

Root-to-Tip Position ( $s$ ): A normalized coordinate where damage is biased toward the tips ( $s \to 1$ ) using a power function $s^{\gamma_s}$ .
Strand Exposure ( $e_i$ ): A per-strand attribute indicating how exposed a strand is to the environment (e.g., outer layers vs. inner layers).
Deterministic Variation ( $n(\xi)$ ): Procedural noise based on persistent strand identifiers and rest-space coordinates to break repetitive patterns.

The field is calculated as:
$P(s, e_i, \xi) = w_s s^{\gamma_s} + w_e e_i + w_n n(\xi)$
$D_{local} = \text{clamp}(D_g \cdot P(s, e_i, \xi), 0, 1)$
Where $D_g$ is a user-specified global damage level. This ensures that all downstream operators share the same spatial support, maintaining correlation between geometric and optical changes.

2. Component Response Operators

The pipeline implements four distinct operators in the GPU shader pipeline, all driven by $D_{local}$ :

Phase-Shift-Based Geometry Disorder:
Instead of uncorrelated white noise (which causes geometric tearing), the method uses a harmonic basis with strand-level phase shifts. The vertex shader displaces strands along a stable rest-space frame:
$V' = V_0 + N(A_{vol} D_{local} s^{\gamma_v} \sin(\omega_v s + 2\pi R_i))$
This preserves the overall curved flow of the hair while introducing frizzy volume expansion, particularly near the tips and boundaries.
Damage-Aware Ombre Discoloration:
The fragment shader interpolates between the base color ( $C_0$ ) and a bleached color ( $C_{bleach}$ ) based on the damage field. The onset of discoloration expands from the tips toward the root as $D_{local}$ increases. The bleached color can be procedurally adjusted for saturation and brightness to simulate melanin loss.
Damage-Conditioned Reflectance Response:
Using a Marschner-based baseline, the method modifies existing scattering lobes rather than introducing new scattering models. $D_{local}$ perturbs the local tangent (decorrelating highlight orientation) and broadens lobe widths while reducing peak weights. This simulates the loss of coherent, glossy highlights.
Procedural Alpha Erosion for Strand Breakage:
To simulate split ends and irregular tearing without modifying geometry, the fragment shader applies high-frequency noise-based alpha erosion. A strand-specific cut threshold ( $C_i$ ) is determined by the tip damage value. Fragments near this threshold are eroded using a combination of a tapering function and high-frequency noise, creating a "frayed" appearance rather than a blunt cut.

3. GPU Pipeline Implementation

The method is implemented entirely in the GPU shader pipeline:

Preprocessing: Asset-dependent quantities (exposure $e_i$ , random seeds, rest-space frames) are precomputed once.
Vertex Shader: Evaluates $D_{local}$ and applies continuous strand deformation (volume expansion).
Fragment Shader: Handles discoloration, reflectance modification, and alpha erosion.
Control: Each component uses a specific response function (e.g., power laws, smoothsteps) to control the onset and intensity of the effect relative to $D_{local}$ , allowing different damage types to manifest at different severity levels.

Key Contributions

Spatial Damage-Field Formulation: A unified field that separates the location of damage from component-specific appearance responses, ensuring spatial correlation between geometry and shading.
Coordinated GPU Pipeline: A real-time implementation that uses the same local field to drive geometric disorder, discoloration, reflectance degradation, and strand-end fragmentation without dense auxiliary geometry.
Evaluation of Scalability and Control: An analysis demonstrating interactive control, predictable scaling with hair complexity, and limited runtime overhead across various hairstyle configurations.

Results and Evaluation

The authors evaluated the method on multiple hairstyles (long, short, wavy) and damage levels.

Qualitative Behavior: The method successfully produces the four target cues (frizz, tip discoloration, degraded highlights, fragmentation) that remain spatially consistent. The damage progression is asset-dependent; for example, long hair shows a clear root-to-tip gradient, while short hair concentrates damage around the silhouette and exposed boundaries.
Damage Progression: As the global damage level ( $D_g$ ) increases, the high-damage region expands from the tips toward the root. Different components respond at different rates: volume expansion and reflectance degradation appear early, while strand breakage becomes prominent only at severe damage levels.
Runtime Performance: On an NVIDIA RTX 3090, the full method increased frame time from 6.59 ms to 7.42 ms (a 12.6% overhead) for a baseline of ~150 FPS. The overhead scales reasonably with strand count (rising from 13.7% at 25k strands to 19.5% at 200k strands), confirming the method's suitability for interactive applications.

Significance and Claims

The paper positions this work as an artist-controllable appearance model rather than a physical predictor of chemical or biological hair damage.

Modest Scope: The authors explicitly state they do not claim physical predictive accuracy regarding BRDF/BSDF data or biological processes. The goal is to reproduce observed visual tendencies (less coherent highlights, tip fraying) in a computationally efficient manner.
Practical Utility: The significance lies in providing a compact, real-time solution that solves the "spatial consistency" problem of damaged hair. By sharing a single control signal, artists can achieve coherent degradation without manually painting separate masks for geometry, color, and opacity.
Future Directions: The authors suggest future work could involve calibrating reflectance parameters with measured data, replacing procedural exposure with image-derived maps, and coupling the appearance field with mechanical properties (stiffness, friction) for damage-aware dynamics.

In conclusion, the paper demonstrates that a shared damage-field approach can effectively model the complex, coupled appearance of severely damaged hair in real-time, offering a practical tool for digital humans, games, and virtual production.