Trimap Classification

A trimap labels each pixel as definite foreground, definite background, or unknown, and this problem asks you to map those labels to a simple alpha value (opacity): 1.0, 0.0, or “unknown”. In image matting, the final goal is to estimate a continuous alpha matte (values in $[0,1]$) that describes how much each pixel belongs to the foreground. A trimap is a coarse prior: the known regions (foreground/background) are fixed, and only the unknown band near boundaries needs more advanced estimation (e.g., closed-form matting, deep matting, etc.).

Formally, matting uses the compositing equation $I = \alpha F + (1-\alpha)B,$ where $I$ is the observed color, $F$ the foreground, $B$ the background, and $\alpha$ the transparency. The trimap constrains this: in known regions, $\alpha$ is exactly 0 or 1; in unknown regions, $\alpha$ must be inferred by a matting algorithm that uses local color statistics, priors, or deep networks. This exercise focuses only on the easy part: translating the trimap label into its corresponding fixed alpha or marking it as unknown.

1. Background Knowledge (key ideas)

• Trimap values and meaning

• 0 → definite background → $\alpha = 0.0$ (fully transparent)

• 255 → definite foreground → $\alpha = 1.0$ (fully opaque)

• 128 → unknown → $\alpha$ must be estimated later

• Why this mapping matters

• Matting algorithms treat known pixels as hard constraints; they never change.

• Only unknown pixels are passed to more complex solvers; this function is part of setting up those constraints cleanly.

2. Algorithm / Approach

This is a simple classification / mapping problem:

• Input: a single trimap pixel value (typically an integer).
• Output:
• 1.0 if the pixel is definite foreground.
• 0.0 if the pixel is definite background.
• None (or similar sentinel, e.g., -1 or NaN depending on the API) if the pixel is in the unknown region.

The general pattern:

• Use conditional checks or a lookup table to map discrete trimap labels to alpha values or “unknown”.
• Do not attempt to estimate alpha for unknown pixels here; just flag them.

3. Step-by-Step Strategy

Assume the function signature is something like: