dichromatic op• Data kinds: rgbimage × labels → vector
• Call: import specularity; specularity.illuminant_from_dichromatic_planes(image_rgb, labels, min_pixels=16, min_plane_ratio=1e-06, min_intersection_ratio=1e-06) (or opsspecular.get("illuminant_from_dichromatic_planes"))
Recover the illuminant colour from two or more materials. → unit 3-vector.
Lee's construction (1986). Under the dichromatic model every pixel of one
material lies in the plane spanned by that material's body colour and the
illuminant colour, so each material contributes a plane through the origin
of RGB, and all of those planes contain the illuminant direction. Two
materials with different body colours therefore intersect in exactly one
line, and that line is the answer — a null-space computation, closed form.
*labels* is an `(H, W)` integer map naming the material at each pixel;
negative labels are ignored (background). Each material needs *min_pixels*
pixels and genuine highlight variation: a material seen with no specular
reflection at all has colours along a single ray, which defines no plane.
That is measured by the second-to-first singular ratio of its colour matrix
and rejected below *min_plane_ratio* rather than contributing an arbitrary
normal.
The returned direction is unit length with a positive component sum
(illuminant colours are positive; the null space fixes the line, not the
sign). On synthetic data with three known materials it reproduces the true
illuminant with a maximum component error of 4.4e-14 and an angular error
that rounds to 0.0 degrees (measured in `tests/test_specularity.py`).
*min_intersection_ratio* guards the answer itself: if the plane normals are
nearly parallel — two materials whose body colours differ only in
brightness, which is the same material twice — the intersection is
ill-conditioned and the call raises instead of returning a direction picked
by rounding error.
Raises `ValueError`: shape or dtype problems as in
:func:specular_diffuse_split; *labels* is not an `(H, W)` integer map
matching the image; more than :data:MAX_MATERIALS distinct labels; fewer
than two materials survive the *min_pixels* and *min_plane_ratio* tests; the
surviving planes do not intersect in a well-determined line.
• specular_photometric family guide
• 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.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• poc_white_balance — py -3.11 examples/poc_white_balance.py
• specular_photometric — py -3.11 examples/specular_photometric.py
vector as input)—
dichromatic)specular_diffuse_split · specular_coefficient_map · specular_free_transform
*Provenance: specularity.py — SPECULAR 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.