app

Focus Stack

Focus Stack fuses at least two focal planes into one all-in-focus image using multilevel Laplacian focus evidence and exports a focus-depth index map.

Requirements And Launch

Inputs should show the same field of view at different focus positions.

labkit_FocusStack_app

Inputs

Add selected image files or choose a folder. Remove blurred, displaced, or otherwise invalid frames before running. The first image defines working geometry; differently sized inputs are resized and the result records how many images required resizing.

Basic Workflow

1. Load at least two focal planes.
2. Remove frames that do not belong to the stack.
3. Start with the Balanced preset.
4. Enable auto-registration when the stack has small rigid camera motion.
5. Tune detail scale, blend radius, and uncertain blend only when artifacts are visible.
6. Run the stack and inspect both fused image and focus map.
7. Export fused PNG, focus-map PNG, and summary CSV.

Fusion Parameters

The public calculation API also accepts minConfidence (default 0.05) and pyramidLevels (default 4, bounded by image geometry).

Algorithm

Images are normalized to grayscale or RGB double data on common geometry. A Gaussian/Laplacian pyramid is built for each plane. Local squared Laplacian detail energy chooses the strongest plane at each level. Confidence is the relative separation between the best and second-best evidence.

High-confidence regions use the winning plane. Low-confidence regions blend normalized evidence weights; zero-evidence regions use equal weights. The weights are spatially smoothed by the level-scaled blend radius and the fused pyramid is reconstructed and clamped to [0,1].

The focus index is the winning input plane at full resolution. Focus coverage is the fraction of pixels assigned to each plane. The algorithm does not correct parallax, deformation, moving objects, illumination changes, or severe registration error.

Outputs

• fused all-in-focus image;
• integer focus-depth index map;
• confidence map and per-plane focus coverage in the run result;
• summary CSV with geometry, parameters, confidence, coverage, registration, and resize information;
• result provenance manifest.

Use Without The GUI

images = {imread("z01.png"), imread("z02.png"), imread("z03.png")};
options = struct("focusWindow", 31, "smoothRadius", 4, ...
    "minConfidence", 0.05, "pyramidLevels", 4);
result = focus_stack.analysisRun.computeFocusStack(images, options);
assert(result.ok, result.message);
imwrite(result.fused, "stacked.png");

Errors And Limitations

• Fewer than two images is an error.
• Reopening a saved stack is strict: an existing image that cannot be decoded aborts restore and leaves the current document unchanged.
• minConfidence must be finite and between 0 and 1.
• Resizing changes sampling and should not substitute for consistent capture.
• Auto-registration should be reviewed; a wrong registration can create a sharp-looking but geometrically false composite.

Related Topics

• Image Measurement family
• Image Library
• API Reference