typed op• Data kinds: lightfield → image
• Call: fullseye.apply(img, "tb_lf_synthetic_aperture", a=0.5, b=0.5) (the 2-D model is one image plus two scalar knobs a,b∈[0,1])
*No figure: this op takes lightfield as input. A Studio program starting from an image cannot reach that type — see the runnable examples below for how it is used.*
Refocus through a shaped aperture — and, with `reduce="median"`, through occluders.
Same shift-and-add geometry as :func:lf_refocus, with two additions:
• *mask* — a `(V, U) weight array from :func:lf_aperture_mask` (or your
own). A small mask is a stopped-down aperture: less defocus blur, less
light. `None` weights every view equally.
• *reduce* — how the aligned views are combined. `mean` is the classical
(and linear) synthetic aperture. `median` is the interesting one: when
a foreground occluder covers a *minority* of the views at a pixel, the
median rejects it and the background behind it is reconstructed —
looking through a fence, or through a rack of parts. `max / min`
are the order-statistic extremes, useful for specular / shadow work.
`median, max and min` use the mask only to *select* views
(weight > 0), because an order statistic has no meaningful weighting;
that is stated here rather than silently ignoring the weights.
The see-through result is not a metaphor. When fewer than half the views
are blocked at a hidden pixel, more than half of them carry the identical
background sample and the median is that sample exactly. Measured
2026-09-01 on a 9x9x64x64 field with an occluder covering 25% of the centre
view at slope 3.0 (blocking at most 46% of the views at any hidden pixel):
RMS against the true, hidden background was 0.0 for `median` and
0.159 for `mean`, with the centre view itself at 0.280. Push the coverage
to 35% (up to 60% of views blocked) and the guarantee is gone — the median
lands at 0.133, worse than nothing.
Returns a `(H, W)` 2-D image.
Raises `ValueError`: *lf* not a valid light field, a *mask* whose
shape is not `(V, U)` or which is non-finite / negative / selects no view,
a non-finite or over-large *slope*, unknown *reduce* / *interp* / *edge*.
Typed bridge of the lightfield op `lf_synthetic_aperture into the 2-D evolution registry: the same implementation, called under the op(v, a, b) convention. This op has no tunable parameter; a and b` are unused.
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• (none yet)
image as input)identity · gaussian · mean_box · bilateral · unsharp · median · min_filter · max_filter
typed)tb_points_to_voxel · tb_estimate_point_normals · tb_iss_keypoints · tb_angle_3points · tb_project_points · tb_render_point_depth · tb_statistical_outlier_removal · tb_radius_outlier_removal
*Provenance: ops.py — 2D operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.