Source code for colour.difference.delta_e

#!/usr/bin/env python
# -*- coding: utf-8 -*-

"""
:math:`\Delta E_{ab}` - Delta E Colour Difference
=================================================

Defines :math:`\Delta E_{ab}` colour difference computation objects:

The following methods are available:

-   :func:`delta_E_CIE_1976`
-   :func:`delta_E_CIE_1994`
-   :func:`delta_E_CIE_2000`
-   :func:`delta_E_CMC`

See Also
--------
`Delta E - Colour Difference IPython Notebook
<http://nbviewer.ipython.org/github/colour-science/colour-ipython/blob/master/notebooks/difference/delta_e.ipynb>`_  # noqa

References
----------
.. [1]  http://en.wikipedia.org/wiki/Color_difference
        (Last accessed 29 August 2014)
"""

from __future__ import division, unicode_literals

import math
import numpy as np

__author__ = 'Colour Developers'
__copyright__ = 'Copyright (C) 2013 - 2014 - Colour Developers'
__license__ = 'New BSD License - http://opensource.org/licenses/BSD-3-Clause'
__maintainer__ = 'Colour Developers'
__email__ = 'colour-science@googlegroups.com'
__status__ = 'Production'

__all__ = ['delta_E_CIE_1976',
           'delta_E_CIE_1994',
           'delta_E_CIE_2000',
           'delta_E_CMC',
           'DELTA_E_METHODS',
           'delta_E']


[docs]def delta_E_CIE_1976(lab1, lab2, **kwargs): """ Returns the difference :math:`\Delta E_{ab}` between two given *CIE Lab* *array_like* colours using *CIE 1976* recommendation. Parameters ---------- lab1 : array_like, (3,) *CIE Lab* *array_like* colour 1. lab2 : array_like, (3,) *CIE Lab* *array_like* colour 2. \*\*kwargs : \*\*, optional Unused parameter provided for signature compatibility with other :math:`\Delta E_{ab}` computation objects. Returns ------- numeric Colour difference :math:`\Delta E_{ab}`. References ---------- .. [2] http://brucelindbloom.com/Eqn_DeltaE_CIE76.html (Last accessed 24 February 2014) Examples -------- >>> lab1 = np.array([100, 21.57210357, 272.2281935]) >>> lab2 = np.array([100, 426.67945353, 72.39590835]) >>> delta_E_CIE_1976(lab1, lab2) # doctest: +ELLIPSIS 451.7133019... """ return np.linalg.norm(np.array(lab1) - np.array(lab2))
[docs]def delta_E_CIE_1994(lab1, lab2, textiles=True, **kwargs): """ Returns the difference :math:`\Delta E_{ab}` between two given *CIE Lab* *array_like* colours using *CIE 1994* recommendation. Parameters ---------- lab1 : array_like, (3,) *CIE Lab* *array_like* colour 1. lab2 : array_like, (3,) *CIE Lab* *array_like* colour 2. textiles : bool, optional Application specific weights. \*\*kwargs : \*\*, optional Unused parameter provided for signature compatibility with other :math:`\Delta E_{ab}` computation objects. Returns ------- numeric Colour difference :math:`\Delta E_{ab}`. References ---------- .. [3] http://brucelindbloom.com/Eqn_DeltaE_CIE94.html (Last accessed 24 February 2014) Examples -------- >>> lab1 = np.array([100, 21.57210357, 272.2281935]) >>> lab2 = np.array([100, 426.67945353, 72.39590835]) >>> delta_E_CIE_1994(lab1, lab2) # doctest: +ELLIPSIS 88.3355530... >>> delta_E_CIE_1994(lab1, lab2, textiles=False) # doctest: +ELLIPSIS 83.7792255... """ k1 = 0.048 if textiles else 0.045 k2 = 0.014 if textiles else 0.015 kL = 2 if textiles else 1 kC = 1 kH = 1 L1, a1, b1 = np.ravel(lab1) L2, a2, b2 = np.ravel(lab2) C1 = math.sqrt(a1 ** 2 + b1 ** 2) C2 = math.sqrt(a2 ** 2 + b2 ** 2) sL = 1 sC = 1 + k1 * C1 sH = 1 + k2 * C1 delta_L = L1 - L2 delta_C = C1 - C2 delta_A = a1 - a2 delta_B = b1 - b2 try: delta_H = math.sqrt(delta_A ** 2 + delta_B ** 2 - delta_C ** 2) except ValueError: delta_H = 0.0 L = (delta_L / (kL * sL)) ** 2 C = (delta_C / (kC * sC)) ** 2 H = (delta_H / (kH * sH)) ** 2 return math.sqrt(L + C + H)
[docs]def delta_E_CIE_2000(lab1, lab2, **kwargs): """ Returns the difference :math:`\Delta E_{ab}` between two given *CIE Lab* *array_like* colours using *CIE 2000* recommendation. Parameters ---------- lab1 : array_like, (3,) *CIE Lab* *array_like* colour 1. lab2 : array_like, (3,) *CIE Lab* *array_like* colour 2. \*\*kwargs : \*\*, optional Unused parameter provided for signature compatibility with other :math:`\Delta E_{ab}` computation objects. Returns ------- numeric Colour difference :math:`\Delta E_{ab}`. References ---------- .. [4] http://brucelindbloom.com/Eqn_DeltaE_CIE2000.html (Last accessed 24 February 2014) Examples -------- >>> lab1 = np.array([100, 21.57210357, 272.2281935]) >>> lab2 = np.array([100, 426.67945353, 72.39590835]) >>> delta_E_CIE_2000(lab1, lab2) # doctest: +ELLIPSIS 94.0356490... """ L1, a1, b1 = np.ravel(lab1) L2, a2, b2 = np.ravel(lab2) kL = 1 kC = 1 kH = 1 l_bar_prime = 0.5 * (L1 + L2) c1 = math.sqrt(a1 * a1 + b1 * b1) c2 = math.sqrt(a2 * a2 + b2 * b2) c_bar = 0.5 * (c1 + c2) c_bar7 = math.pow(c_bar, 7) g = 0.5 * (1 - math.sqrt(c_bar7 / (c_bar7 + 25 ** 7))) a1_prime = a1 * (1 + g) a2_prime = a2 * (1 + g) c1_prime = math.sqrt(a1_prime * a1_prime + b1 * b1) c2_prime = math.sqrt(a2_prime * a2_prime + b2 * b2) c_bar_prime = 0.5 * (c1_prime + c2_prime) h1_prime = (math.atan2(b1, a1_prime) * 180) / math.pi if h1_prime < 0: h1_prime += 360 h2_prime = (math.atan2(b2, a2_prime) * 180) / math.pi if h2_prime < 0.0: h2_prime += 360 h_bar_prime = (0.5 * (h1_prime + h2_prime + 360) if math.fabs(h1_prime - h2_prime) > 180 else 0.5 * (h1_prime + h2_prime)) t = (1 - 0.17 * math.cos(math.pi * (h_bar_prime - 30) / 180) + 0.24 * math.cos(math.pi * (2 * h_bar_prime) / 180) + 0.32 * math.cos(math.pi * (3 * h_bar_prime + 6) / 180) - 0.20 * math.cos(math.pi * (4 * h_bar_prime - 63) / 180)) if math.fabs(h2_prime - h1_prime) <= 180: delta_h_prime = h2_prime - h1_prime else: delta_h_prime = (h2_prime - h1_prime + 360 if h2_prime <= h1_prime else h2_prime - h1_prime - 360) delta_L_prime = L2 - L1 delta_C_prime = c2_prime - c1_prime delta_H_prime = (2 * math.sqrt(c1_prime * c2_prime) * math.sin(math.pi * (0.5 * delta_h_prime) / 180)) sL = 1 + ((0.015 * (l_bar_prime - 50) * (l_bar_prime - 50)) / math.sqrt(20 + (l_bar_prime - 50) * (l_bar_prime - 50))) sC = 1 + 0.045 * c_bar_prime sH = 1 + 0.015 * c_bar_prime * t delta_theta = (30 * math.exp(-((h_bar_prime - 275) / 25) * ((h_bar_prime - 275) / 25))) c_bar_prime7 = c_bar_prime ** 7 rC = math.sqrt(c_bar_prime7 / (c_bar_prime7 + 25 ** 7)) rT = -2 * rC * math.sin(math.pi * (2 * delta_theta) / 180) return math.sqrt( (delta_L_prime / (kL * sL)) * (delta_L_prime / (kL * sL)) + (delta_C_prime / (kC * sC)) * (delta_C_prime / (kC * sC)) + (delta_H_prime / (kH * sH)) * (delta_H_prime / (kH * sH)) + (delta_C_prime / (kC * sC)) * (delta_H_prime / (kH * sH)) * rT)
[docs]def delta_E_CMC(lab1, lab2, l=2, c=1): """ Returns the difference :math:`\Delta E_{ab}` between two given *CIE Lab* *array_like* colours using *Colour Measurement Committee* recommendation. The quasimetric has two parameters: *Lightness* (l) and *chroma* (c), allowing the users to weight the difference based on the ratio of l:c. Commonly used values are 2:1 for acceptability and 1:1 for the threshold of imperceptibility. Parameters ---------- lab1 : array_like, (3,) *CIE Lab* *array_like* colour 1. lab2 : array_like, (3,) *CIE Lab* *array_like* colour 2. l : numeric, optional Lightness weighting factor. c : numeric, optional Chroma weighting factor. Returns ------- numeric Colour difference :math:`\Delta E_{ab}`. References ---------- .. [5] http://brucelindbloom.com/Eqn_DeltaE_CMC.html (Last accessed 24 February 2014) Examples -------- >>> lab1 = np.array([100, 21.57210357, 272.2281935]) >>> lab2 = np.array([100, 426.67945353, 72.39590835]) >>> delta_E_CMC(lab1, lab2) # doctest: +ELLIPSIS 172.7047712... """ L1, a1, b1 = np.ravel(lab1) L2, a2, b2 = np.ravel(lab2) c1 = math.sqrt(a1 * a1 + b1 * b1) c2 = math.sqrt(a2 * a2 + b2 * b2) sl = 0.511 if L1 < 16 else (0.040975 * L1) / (1 + 0.01765 * L1) sc = 0.0638 * c1 / (1 + 0.0131 * c1) + 0.638 h1 = 0 if c1 < 0.000001 else (math.atan2(b1, a1) * 180) / math.pi while h1 < 0: h1 += 360 while h1 >= 360: h1 -= 360 t = (0.56 + math.fabs(0.2 * math.cos((math.pi * (h1 + 168)) / 180)) if 164 <= h1 <= 345 else 0.36 + math.fabs(0.4 * math.cos((math.pi * (h1 + 35)) / 180))) c4 = c1 * c1 * c1 * c1 f = math.sqrt(c4 / (c4 + 1900)) sh = sc * (f * t + 1 - f) delta_L = L1 - L2 delta_C = c1 - c2 delta_A = a1 - a2 delta_B = b1 - b2 delta_H2 = delta_A * delta_A + delta_B * delta_B - delta_C * delta_C v1 = delta_L / (l * sl) v2 = delta_C / (c * sc) v3 = sh return math.sqrt(v1 * v1 + v2 * v2 + (delta_H2 / (v3 * v3)))
DELTA_E_METHODS = { 'CIE 1976': delta_E_CIE_1976, 'CIE 1994': delta_E_CIE_1994, 'CIE 2000': delta_E_CIE_2000, 'CMC': delta_E_CMC, } """ Supported *Delta E* computations methods. DELTA_E_METHODS : dict ('CIE 1976', 'CIE 1994', 'CIE 2000', 'CMC') Aliases: - 'cie1976': 'CIE 1976' - 'cie1994': 'CIE 1994' - 'cie2000': 'CIE 2000' """ DELTA_E_METHODS['cie1976'] = DELTA_E_METHODS['CIE 1976'] DELTA_E_METHODS['cie1994'] = DELTA_E_METHODS['CIE 1994'] DELTA_E_METHODS['cie2000'] = DELTA_E_METHODS['CIE 2000']
[docs]def delta_E(lab1, lab2, method='CMC', **kwargs): """ Returns the *Lightness* :math:`L^*` using given method. Parameters ---------- lab1 : array_like, (3,) *CIE Lab* *array_like* colour 1. lab2 : array_like, (3,) *CIE Lab* *array_like* colour 2. method : unicode, optional ('CIE 1976', 'CIE 1994', 'CIE 2000', 'CMC') Computation method. \*\*kwargs : \*\* Keywords arguments. Returns ------- numeric Colour difference :math:`\Delta E_{ab}`. Examples -------- >>> lab1 = np.array([100, 21.57210357, 272.2281935]) >>> lab2 = np.array([100, 426.67945353, 72.39590835]) >>> delta_E(lab1, lab2) # doctest: +ELLIPSIS 172.7047712... >>> delta_E(lab1, lab2, method='CIE 1976') # doctest: +ELLIPSIS 451.7133019... >>> delta_E(lab1, lab2, method='CIE 1994') # doctest: +ELLIPSIS 88.3355530... >>> delta_E(lab1, lab2, method='CIE 1994', textiles=False) # noqa # doctest: +ELLIPSIS 83.7792255... >>> delta_E(lab1, lab2, method='CIE 2000') # doctest: +ELLIPSIS 94.0356490... """ return DELTA_E_METHODS.get(method)(lab1, lab2, **kwargs)