Alpha-Gated Edge Despill (AGED)

Zero-parameter, Alpha-aware chroma key spill suppression for real-time compositing.

✨ Visual Results

Traditional Global Despill (Note damaged interior colors)

AGED Edge-Only Despill (Perfect interior preservation)

Note: AGED only processes pixels with 0 < α < 1, leaving fully opaque areas completely untouched.

AGED is a lightweight, high-performance despill algorithm specifically designed for alpha boundary pixels (0 < α < 1). Unlike traditional global despill methods, AGED applies conditional channel replacement only to edge pixels where green spill actually occurs during compositing, preserving interior colors perfectly.

Further reading:

• Algorithm details
• Comparison with traditional despill

Key Features

• 🎯 Alpha-Gated Processing: Only processes pixels with 0 < α < 1, avoiding interior color distortion.
• ⚡ Zero Parameters: No manual tuning of spillmix, thresholds, or color spaces.
• 🧮 Hardware-Friendly: Branchless SIMD implementation, ideal for GPU/FPGA.
• 🎨 Perceptually Optimized: Hard-decision replacement eliminates green fringing without complex luminance compensation.
• 📹 Real-Time: O(N) complexity, 4K@60fps on consumer GPUs.

Mathematical Formulation

Given foreground pixel F = (R, G, B), background B, and alpha mask α ∈ [0,1]:

1. Edge Detection Mask

$$\mathcal{M}_{\text{edge}} = \mathbb{I}(\alpha \in (0,1))$$

2. Spill Condition (Green Screen)

$$\mathcal{M}_{\text{spill}} = \mathbb{I}(G = \max(R, G, B))$$

3. Conditional Replacement

$$F' = \begin{cases} (R, \max(R, B), B) &amp; \text{if } \mathcal{M}_{\text{edge}} \land \mathcal{M}_{\text{spill}} \ F &amp; \text{otherwise} \end{cases}$$

4. Standard Alpha Compositing

$$C = \alpha \cdot F' + (1-\alpha) \cdot B_{\text{bg}}$$

Quick Start

Python (NumPy/Cython)

import aged
import cv2

# Load your RGBA image (H, W, 4)
rgba = cv2.imread("keyed_image.png", cv2.IMREAD_UNCHANGED) / 255.0
alpha = rgba[:, :, 3]

# Apply AGED despill (Maintains float32 dtype automatically)
result = aged.despill(rgba[:, :, :3], alpha, screen_type='green')

GLSL (Shader)

#version 330 core

// Uniforms
uniform sampler2D foregroundTex;
uniform sampler2D backgroundTex;
uniform int screenType; // 0 for Green, 1 for Blue

// Inputs and Outputs
in vec2 TexCoord;
out vec4 FragColor;

vec3 aged_despill(vec3 color, float alpha, int type) {
    // Alpha gating: early exit for opaque/interior pixels
    if (alpha <= 0.001 || alpha >= 0.999) {
        return color;
    }
    
    vec3 result = color;
    
    if (type == 0) { // Green screen
        if (color.g > color.r && color.g > color.b) {
            result.g = max(color.r, color.b);
        }
    } else { // Blue screen
        if (color.b > color.r && color.b > color.g) {
            result.b = max(color.r, color.g);
        }
    }
    
    return result;
}

void main() {
    vec4 fg = texture(foregroundTex, TexCoord);
    vec3 bg = texture(backgroundTex, TexCoord).rgb;
    
    // Apply AGED before compositing (assumes straight alpha)
    vec3 despilled = aged_despill(fg.rgb, fg.a, screenType);
    
    // Standard alpha composite
    vec3 finalColor = mix(bg, despilled, fg.a);
    
    FragColor = vec4(finalColor, 1.0);
}

Why AGED vs Traditional Methods?

Problem with Global Despill

Traditional algorithms process the entire image:

• ❌ Damages legitimate greens in foreground (clothing, objects).
• ❌ Requires manual tuning of spillmix parameters.
• ❌ Computational waste on opaque interior pixels.

AGED Solution

• ✅ Surgical precision: Only touches semi-transparent edge pixels (0 < α < 1).
• ✅ Zero-config: Hard decision based on max channel eliminates tuning.
• ✅ Cache-friendly: Sequential memory access pattern for edge pixels only.

Algorithm Comparison

Method	Parameters	Edge Awareness	Interior Preservation	Complexity
AGED (Ours)	0	✅ Yes	✅ Perfect	O(N)
Global Despill	1+	❌ No	❌ No	O(N)
Primatte	10+	⚠️ Partial	⚠️ Partial	O(N log N)
LWGSS (AI)	Trainable	✅ Yes	✅ Yes	O(N · C)

Limitations & Future Work

• Premultiplied Alpha: AGED assumes unassociated (straight) alpha. If using premultiplied alpha, foreground channels must be un-premultiplied prior to applying AGED to ensure accurate green screen comparisons.
• Luminance Shift: Replacement may darken/brighten edges slightly (acceptable for most broadcast scenarios).
• Secondary Spill: Does not handle global secondary spill (e.g., green reflecting onto skin from clothing far from the edge).

Citation

If you use AGED in your research or production:

@software{aged_despill,
  author = {Your Name},
  title = {Alpha-Gated Edge Despill (AGED): Zero-Parameter Chroma Key Optimization},
  year = {2026},
  url = {[https://github.com/yourusername/aged-despill](https://github.com/yourusername/aged-despill)}
}

License

MIT License - See LICENSE file.