Solve Matting Equation for Alpha

mediumPROImage MattingMattingCV: Computational Photography

Solve the matting equation for unknown alpha given observed color and known foreground/background.

The matting equation states that an observed pixel is a blend of foreground and background:

$I = \alpha F + (1 - \alpha) B$

If we know F (foreground color) and B (background color), we can solve for alpha:

$\alpha = \frac{I - B}{F - B}$

In practice, we use the channel with the largest |F - B| difference for numerical stability, as channels where F ≈ B provide unreliable estimates.

This is used in chroma keying (green screen) where F is known (subject) and B is known (green screen).

Example

Input

solve_alpha((128, 50, 50), (255, 100, 100), (0, 0, 0))

Output

0.502

Reasoning

Finding alpha from RGB channels: Channel differences |F - B|:

R: |255 - 0| = 255 (largest - use this)
G: |100 - 0| = 100
B: |100 - 0| = 100

Using red channel: α = (128 - 0) / (255 - 0) = 128/255 ≈ 0.502

Constraints

observed, foreground, background: RGB tuples [0, 255]
Return alpha clamped to [0, 1] and rounded to 4 decimal places
Use the channel with largest |F - B| for best accuracy

Learn this in

Computational PhotographyCV Study Plan

More from CV: Computational Photography

Trimap-Based Alpha Estimationhard Drago Adaptive Log Mappinghard Gram Matrix for Stylehard Alpha Matte Compositingmedium Reinhard Global Tone Mappingmedium

Concept primerPREVIEW

Image Matting: Alpha Estimation from the Matting Equation

Background Knowledge

The Matting Equation and Compositing Model

The fundamental principle underlying image matting is the compositing equation, which models how foreground and background colors blend to create an observed pixel. The equation $I =$ \alpha $F + (1 -$ \alpha $) B$ represents a linear blend where:

$I$ is the observed pixel color in the composite image
$\alpha$ (alpha matte) is the foreground opacity, ranging from 0 (fully transparent/background) to 1 (fully opaque/foreground)
$F$ is the foreground color
$B$ is the background color

This model assumes that each color channel combines independently, making it possible to solve for alpha algebraically when foreground and background colors are known. The inverse problem—recovering alpha from observed pixels when F and B are known—is the core challenge in image matting, which refers to the interactive process of estimating foreground opacity and extracting foreground layers.

Numerical Stability and Channel Selection

A critical practical consideration is numerical stability. When solving $\alpha = \frac{I - B}{F - B}$ , the denominator $(F - B)$ becomes problematic when foreground and background colors are similar in a particular channel. Dividing by a small number amplifies noise and produces unreliable alpha estimates. Therefore, the algorithm should select the color channel where $|F - B|$ is largest—this channel provides the most discriminative information and the most stable division. This is particularly important in applications like chroma keying (green screen removal), where the background is a known, uniform color and the foreground subject is extracted by solving for alpha across all pixels.

Algorithm/Approach

The general approach to solving the matting equation involves three main steps:

Input Validation: Verify that foreground and background colors are sufficiently different to enable stable computation
Channel Selection: Identify the color channel with maximum $|F - B|$ difference
Alpha Computation: Apply the inverse matting equation using the selected channel

This approach leverages the linearity of the compositing model and treats each pixel independently, making it straightforward to vectorize across entire images.

Step-by-Step Strategy

PRO

Continue the full explanation

You're reading the free preview. Unlock the complete walkthrough, the code editor, test runner and reference solution with Premium.

PRO

Editor locked

The code editor is locked for Pro problems. It is only available for free problems. Please upgrade to gain access to the code editor for all problems.

solution.py

Python 3.13

Test results