deltaE_ciede2000 crashes on mismatched-ndim inputs (e.g. image vs single reference color); sibling functions work fine #8326

Description

While testing broadcasting behavior across the deltaE_* functions, I ran into a crash in deltaE_ciede2000 when lab1 and lab2 don't have the same number of dimensions — for example, comparing a whole image of colors against a single reference color. The other functions (deltaE_cie76, deltaE_ciede94, deltaE_cmc) handle this fine, so it looks specific to deltaE_ciede2000.

This isn't limited to the dev _skimage2 module — the same buggy pattern is present in the current released skimage.color.deltaE_ciede2000 as well, so this affects real users on the latest stable release too.

What's going on:

channel_axis = channel_axis % lab1.ndim

This normalizes channel_axis using only lab1's number of dimensions, and then the same value gets reused for lab2 too when calling np.moveaxis a few lines down — which breaks if lab2 doesn't have that many dimensions to begin with.

The other three functions don't do this pre-normalization step at all — they just pass channel_axis straight into np.moveaxis, and numpy normalizes it separately for each array based on its own shape. That's why they don't have this problem.

Worth noting: the original PR that added these functions (#665) explicitly says broadcasting mismatched shapes should work (x.shape == (3,) vs y.shape == (10, 10, 3)), so this looks like an unintentional regression introduced when channel_axis was added later, not a deliberate restriction.

I already have a fix worked out (normalize the axis separately for each array instead of sharing one value) and it passes against this case, a few other shape combos, and the existing 29 tests in test_delta_e.py. Happy to open a PR if that sounds good.

Way to reproduce

import numpy as np
from skimage.color import deltaE_cie76, deltaE_ciede94, deltaE_ciede2000

lab1 = np.array([[50., 20., 30.], [60., -10., 15.]])  # image: 2 colors, shape (2, 3)
lab2 = np.array([55., 18., 28.])                        # single reference color, shape (3,)

deltaE_cie76(lab1, lab2)     # works fine -> array([ 5.74456265, 31.27299154])
deltaE_ciede94(lab1, lab2)   # works fine -> array([ 5.12397703, 23.28484273])
deltaE_ciede2000(lab1, lab2) # AxisError: source: axis 1 is out of bounds for array of dimension 1

Swapping the order (lab1 as the single color, lab2 as the image) fails too, but with a different error: ValueError: not enough values to unpack (expected 3, got 2).

Version information

3.14.4 (main, Jun 18 2026, 14:25:02) [GCC 15.2.0]
scikit-image version: 0.26.1rc0.dev0+git20260909.b6e9c5ce5
numpy version: 2.5.3
scipy version: 1.18.1