ciecam02

An easy fast transformer between rgb and ciecam02 color space.

Converts color between rgb and a simplified form of the CIECAM02 model named JCH, wrapped in the numpy arrays in a quick algorithm, for a practical use for doing algorithms of images, displaying, etc.

introduce

Munsell color system

CIECAM02 approximately linearize Moncell color system.

Color type rgb could be use widely in display devices and image formats. The data form is an integer list [r, g, b], where r g b is among 0 - 255.

CIECAM02 produces multiple correlates, like H, J, z, Q, t, C, M, s. Some of them represent similar concepts, such as C means chroma and M colorfulness s saturation correlate the same thing in different density. We need only 3 major property of these arguments to completely represent a color, and we can get other properties or reverse algorithms.

Color type jch is a float list like [j, c, h], where 0.0 < j < 100.0, 0.0 < h < 360.0, and 0.0 < c. the max value of c does not limit, and may produce exceeds when transform to rgb. The effective value of max c varies. Probablly for red color h 0.0, and brightness j 50.0, c reach the valid maximum, values about 160.0.

And jch comes from the CIECAM02 model outputs as an float list like [j, c, h], and some distortion was made to obtain a proper proportion.

j values the same as J, the brightness.

c values the same as C, the chroma.

h compress the original H from 0-400 to 0-360 by simply * 0.9 for representing in a polar coordinates.

visual environment variable

Affections was considered in the ciecam02 model and several mesurement was constructed. We just deal them as constants.

whitepoint = [95.05, 100.00, 108.88],
env = [1.0, 0.69, 1.0],
lightindensity = 80
bgindensity = 16

And you can config these values to simulate a different environment.

implementation

Implementation relies on numpy, which acts in a high performance of transformation. Process large image data as array at once.

install

pip install ciecam02

Usage

Basic functions:

import numpy as np
from ciecam02 import rgb2jch, jch2rgb
rgb = np.array([[56, 34, 199],
                [255, 255, 255]
                ])

rgb2jch(rgb)

Out[]: array([[ 21.4432157,  74.80048318, 284.3167947 ],
              [ 99.99968129,  1.49090566, 242.41103965]])

jch2rgb(rgb2jch(rgb))

Out[]:
    array([[ 56,  34, 199],
           [255, 255, 255]], dtype=uint8)

For image analyzing, convert image rgb data to np.array flattened to a 2d array, then apply the transform:

import numpy as np
from ciecam02 import rgb2jch, jch2rgb
from PIL import Image

im = Image.open("image.png")
rgb = np.array(im)
shape = rgb.shape
jch = rgb2jch(rgb.reshape(-1, 3))

# do the work.
# ...   

# reverse back to a new image
rgb = jch2rgb(jch).reshape(shape)
im = Image.fromarray(rgb)
im.show()

When doing the reverse transform, rgb values may exceeds. jch2rgb control values in the threshhold of 255. if you want to see those points who are out of range, use the following functions:

import numpy as np 
from ciecam02 import jch2xyz, xyz2infinitergb

...

xyz = jch2xyz(jch)
inf_rgb = xyz2infinitergb(rgb)

# then select the exceeded values for your purpose.

functions

The procedures rely on an intermediate representation of xyz color space. Forward and reverse transform first comes to xyz, then do transformation to the target space.

• rgb2xyz(rgb) :

  returns an xyz 2d array astype float shapes [[x, y, z], ...].

• xyz2rgb(xyz) :

  returns an rgb array astype uint8. coud convert to Image object by Image.fromarray(rgb)

• xyz2infinitergb(xyz) :

  returns an rgb array with exceeded values in a float astype. if you want transform to an image object, you should convert to a uint8 first.

• xyz2cam02(xyz) :

  returns the full property of features that cam02 solves, with data undistorted. [[h, H, J, Q, C, M, s],...]

• rgb2jch(color):

  returns a jch array. It is a compound of 'xyz2cam02(rgb2xyz)', then select J, C, H compressed out in a float type.

• jch2xyz(jch):

  returns an xyz array in a float type.

• jch2rgb(jch):

  returns rgb array astype uint8. Compounds of xyz2rgb(jch2xyz(jch)).

config

predefined values like below:

whitepoint = {'white': [95.05, 100.00, 108.88],
              'c': [109.85, 100.0, 35.58]}

env = {'dim': [0.9, 0.59, 0.9],
       'average': [1.0, 0.69, 1.0],
       'dark': [0.8, 0.525, 0.8]}
lightindensity = {'default': 80.0, 'high': 318.31, 'low': 31.83}
bgindensity = {'default': 16.0, 'high': 20.0, 'low': 10.0}

You can config them by set_config(whitepoint, env, light, bg), from the dict keys in each constant:

from ciecam02 import set_config
set_config('c', 'dim', 'low', 'high')

Or you can directly change the global constant, they were predefined like this:

currentwhite = whitepoint['white']
currentenv = env['average']
currentlight = lightindensity['default']
currentbg = bgindensity['default']

Modify for a custom presets:

currentwhite = [101.0, 100.0, 95]
currentenv = [0.96, 0.59, 0.99]
currentlight = 200.0
currentbg = 18.0

Model appearence

See some difference from Moncell and CAM02.

Comparisons

CIECAM02:

Moncell:

CIECAM02:

Moncell:

Dimensions

Lisence

MIT Lisence.