typed op• Data kinds: qimage → qimage
• Call: fullseye.apply(img, "tb_quat_color_filter", a=0.5, b=0.5) (the 2-D model is one image plus two scalar knobs a,b∈[0,1])
Keep or remove one colour direction, exactly. → (H, W, 4).
`mode="remove" returns v - (v.g) g for the unit RGB direction g`:
the component along `g` is exactly zero everywhere afterwards, to
machine precision (measured max residual 5.8e-16 and 6.5e-16 on two random
colour images — seed-dependent only at the 1e-16 level),
and `remove + keep` reproduces the input to 0.0 exactly.
`mode="keep" returns the complementary (v.g) g`. The scalar part is
passed through untouched in both.
There is no default mode. The two are opposites, both return a valid
picture, and neither raises — so choosing for the caller would be a coin flip
that never announces itself.
Not a new algorithm, and this docstring will not pretend otherwise
-----------------------------------------------------------------
The `remove` branch is the specular-invariant projection of Mallick et
al. (2005), which this repository already implements as
`specularity.specular_free_transform for the rgbimage` sort. Rather
than write the same three lines twice, this operator *delegates* to it — so
agreement between the two sorts is by construction rather than by luck, and a
future fix in one is a fix in both. What is added here is the `keep`
branch (which has no counterpart there) and the `qimage` sort, so the
projection composes with :func:quat_color_rotate and :func:qft2.
What this can do that a channelwise pipeline cannot
---------------------------------------------------
A per-channel filter applies a diagonal matrix, and `I - g g^T` is diagonal
only when `g is a coordinate axis. For g = (1,1,1)/sqrt(3)` — remove
the grey axis, i.e. keep only chromatic content — the *best possible*
diagonal approximation is off by `||P - diag(P)||_2 = 0.666667` in operator
norm. Concretely, a pure red pixel `(1, 0, 0)` must become
`(0.666667, -0.333333, -0.333333)`; the best diagonal filter can only reach
`(0.666667, 0, 0), an error of 0.471405` — it cannot put anything into
the green and blue channels, because it multiplies each channel by a number
and both start at zero. The impossibility is structural, not a tuning gap.
(A full 3x3 colour matrix, of course, does it exactly; see
:func:quat_color_rotate for that half of the accounting.)
Raises `ValueError: *qimage* is not a valid (H, W, 4)` field;
*direction_rgb* is not a finite non-zero 3-vector; *mode* is not
`'remove' / 'keep'`.
Typed bridge of the quat op `quat_color_filter 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)
qimage as input)identity · tb_quaternion_to_rgb · tb_quat_norm · tb_quat_conjugate_image · tb_quat_normalize_image · tb_monogenic_amplitude · tb_monogenic_phase · tb_monogenic_orientation
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.