Source code for psychopy.tools.colorspacetools

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

# Part of the PsychoPy library
# Copyright (C) 2002-2018 Jonathan Peirce (C) 2019-2021 Open Science Tools Ltd.
# Distributed under the terms of the GNU General Public License (GPL).

"""Tools related to working with various color spaces.

The routines provided in the module are used to transform color coordinates
between spaces. Most of the functions here are *vectorized*, allowing for array
inputs to convert multiple color values at once.

**As of version 2021.0 of PsychoPy**, users ought to use the
:class:`~psychopy.colors.Color` class for working with color coordinate values.

"""
from __future__ import absolute_import, division, print_function

__all__ = ['srgbTF', 'rec709TF', 'cielab2rgb', 'cielch2rgb', 'dkl2rgb',
           'dklCart2rgb', 'rgb2dklCart', 'hsv2rgb', 'rgb2lms', 'lms2rgb',
           'rgb2hsv', 'rescaleColor']

from past.utils import old_div
import numpy
from psychopy import logging
from psychopy.tools.coordinatetools import sph2cart


def unpackColors(colors):  # used internally, not exported by __all__
    """Reshape an array of color values to Nx3 format.

    Many color conversion routines operate on color data in Nx3 format, where
    rows are color space coordinates. 1x3 and NxNx3 input are converted to Nx3
    format. The original shape and dimensions are also returned, allowing the
    color values to be returned to their original format using 'reshape'.

    Parameters
    ----------
    colors : ndarray, list or tuple of floats
        Nx3 or NxNx3 array of colors, last dim must be size == 3 specifying each
        color coordinate.

    Returns
    -------
    tuple
        Nx3 ndarray of converted colors, original shape, original dims.

    """
    # handle the various data types and shapes we might get as input
    colors = numpy.asarray(colors, dtype=float)

    orig_shape = colors.shape
    orig_dim = colors.ndim
    if orig_dim == 1 and orig_shape[0] == 3:
        colors = numpy.array(colors, ndmin=2)
    elif orig_dim == 2 and orig_shape[1] == 3:
        pass  # NOP, already in correct format
    elif orig_dim == 3 and orig_shape[2] == 3:
        colors = numpy.reshape(colors, (-1, 3))
    else:
        raise ValueError(
            "Invalid input dimensions or shape for input colors.")

    return colors, orig_shape, orig_dim


[docs]def rescaleColor(rgb, convertTo='signed', clip=False): """Rescale RGB colors. This function can be used to convert RGB value triplets from the PsychoPy signed color format to the normalized OpenGL format. PsychoPy represents colors using values between -1 and 1. However, colors are commonly represented using values between 0 and 1 when working with OpenGL and various other contexts. This function simply rescales values to switch between these formats. Parameters ---------- rgb : `array_like` 1-, 2-, 3-D vector of RGB coordinates to convert. The last dimension should be length-3 in all cases, specifying a single coordinate. convertTo : `str` If 'signed', this function will assume `rgb` is in OpenGL format [0:1] and rescale them to PsychoPy's format [-1:1]. If 'unsigned', input values are treated as OpenGL format and will be rescaled to use PsychoPy's. Default is 'signed'. clip : bool Clip values to the range that can be represented on a display. This is an optional step. Default is `False`. Returns ------- ndarray Rescaled values with the same shape as `rgb`. Notes ----- The `convertTo` argument also accepts strings 'opengl' and 'psychopy' as substitutes for 'signed' and 'unsigned', respectively. This might be more explicit in some contexts. Examples -------- Convert a signed RGB value to unsigned format:: rgb_signed = [-1, 0, 1] rgb_unsigned = rescaleColor(rgb_signed, convertTo='unsigned') """ # While pretty simple, this operation is done often enough to justify having # its own function to avoid writing it out all the time. It also explicitly # shows the direction of which values are being rescaled to make code more # readable. if convertTo == 'signed' or convertTo == 'psychopy': rgb_out = rgb * 2 - 1 # from OpenGL to PsychoPy format elif convertTo == 'unsigned' or convertTo == 'opengl': rgb_out = (rgb + 1) / 2. # from PsychoPy to OpenGL else: raise ValueError("Invalid value for `convertTo`, can either be " "'signed' or 'unsigned'.") if clip: rgb_out = numpy.clip(rgb_out, -1 if convertTo == 'signed' else 0, 1) return rgb_out
[docs]def srgbTF(rgb, reverse=False, **kwargs): """Apply sRGB transfer function (or gamma) to linear RGB values. Input values must have been transformed using a conversion matrix derived from sRGB primaries relative to D65. Parameters ---------- rgb : tuple, list or ndarray of floats Nx3 or NxNx3 array of linear RGB values, last dim must be size == 3 specifying RBG values. reverse : boolean If True, the reverse transfer function will convert sRGB -> linear RGB. Returns ------- ndarray Array of transformed colors with same shape as input. """ rgb, orig_shape, orig_dim = unpackColors(rgb) # apply the sRGB TF if not reverse: # applies the sRGB transfer function (linear RGB -> sRGB) to_return = numpy.where( rgb <= 0.0031308, rgb * 12.92, (1.0 + 0.055) * rgb ** (1.0 / 2.4) - 0.055) else: # do the inverse (sRGB -> linear RGB) to_return = numpy.where( rgb <= 0.04045, rgb / 12.92, ((rgb + 0.055) / 1.055) ** 2.4) if orig_dim == 1: to_return = to_return[0] elif orig_dim == 3: to_return = numpy.reshape(to_return, orig_shape) return to_return
[docs]def rec709TF(rgb, **kwargs): """Apply the Rec 709 transfer function (or gamma) to linear RGB values. This transfer function is defined in the ITU-R BT.709 (2015) recommendation document (http://www.itu.int/rec/R-REC-BT.709-6-201506-I/en) and is commonly used with HDTV televisions. Parameters ---------- rgb : tuple, list or ndarray of floats Nx3 or NxNx3 array of linear RGB values, last dim must be size == 3 specifying RBG values. Returns ------- ndarray Array of transformed colors with same shape as input. """ rgb, orig_shape, orig_dim = unpackColors(rgb) # applies the Rec.709 transfer function (linear RGB -> Rec.709 RGB) # mdc - I didn't compute the inverse for this one. to_return = numpy.where(rgb >= 0.018, 1.099 * rgb ** 0.45 - 0.099, 4.5 * rgb) if orig_dim == 1: to_return = to_return[0] elif orig_dim == 3: to_return = numpy.reshape(to_return, orig_shape) return to_return
[docs]def cielab2rgb(lab, whiteXYZ=None, conversionMatrix=None, transferFunc=None, clip=False, **kwargs): """Transform CIE L*a*b* (1976) color space coordinates to RGB tristimulus values. CIE L*a*b* are first transformed into CIE XYZ (1931) color space, then the RGB conversion is applied. By default, the sRGB conversion matrix is used with a reference D65 white point. You may specify your own RGB conversion matrix and white point (in CIE XYZ) appropriate for your display. Parameters ---------- lab : tuple, list or ndarray 1-, 2-, 3-D vector of CIE L*a*b* coordinates to convert. The last dimension should be length-3 in all cases specifying a single coordinate. whiteXYZ : tuple, list or ndarray 1-D vector coordinate of the white point in CIE-XYZ color space. Must be the same white point needed by the conversion matrix. The default white point is D65 if None is specified, defined as X, Y, Z = 0.9505, 1.0000, 1.0890. conversionMatrix : tuple, list or ndarray 3x3 conversion matrix to transform CIE-XYZ to RGB values. The default matrix is sRGB with a D65 white point if None is specified. Note that values must be gamma corrected to appear correctly according to the sRGB standard. transferFunc : pyfunc or None Signature of the transfer function to use. If None, values are kept as linear RGB (it's assumed your display is gamma corrected via the hardware CLUT). The TF must be appropriate for the conversion matrix supplied (default is sRGB). Additional arguments to 'transferFunc' can be passed by specifying them as keyword arguments. Gamma functions that come with PsychoPy are 'srgbTF' and 'rec709TF', see their docs for more information. clip : bool Make all output values representable by the display. However, colors outside of the display's gamut may not be valid! Returns ------- ndarray Array of RGB tristimulus values. Example ------- Converting a CIE L*a*b* color to linear RGB:: import psychopy.tools.colorspacetools as cst cielabColor = (53.0, -20.0, 0.0) # greenish color (L*, a*, b*) rgbColor = cst.cielab2rgb(cielabColor) Using a transfer function to convert to sRGB:: rgbColor = cst.cielab2rgb(cielabColor, transferFunc=cst.srgbTF) """ lab, orig_shape, orig_dim = unpackColors(lab) if conversionMatrix is None: # XYZ -> sRGB conversion matrix, assumes D65 white point # mdc - computed using makeXYZ2RGB with sRGB primaries conversionMatrix = numpy.asmatrix([ [3.24096994, -1.53738318, -0.49861076], [-0.96924364, 1.8759675, 0.04155506], [0.05563008, -0.20397696, 1.05697151] ]) if whiteXYZ is None: # D65 white point in CIE-XYZ color space # See: https://en.wikipedia.org/wiki/SRGB whiteXYZ = numpy.asarray([0.9505, 1.0000, 1.0890]) L = lab[:, 0] # lightness a = lab[:, 1] # green (-) <-> red (+) b = lab[:, 2] # blue (-) <-> yellow (+) wht_x, wht_y, wht_z = whiteXYZ # white point in CIE-XYZ color space # convert Lab to CIE-XYZ color space # uses reverse transformation found here: # https://en.wikipedia.org/wiki/Lab_color_space xyz_array = numpy.zeros(lab.shape) s = (L + 16.0) / 116.0 xyz_array[:, 0] = s + (a / 500.0) xyz_array[:, 1] = s xyz_array[:, 2] = s - (b / 200.0) # evaluate the inverse f-function delta = 6.0 / 29.0 xyz_array = numpy.where(xyz_array > delta, xyz_array ** 3.0, (xyz_array - (4.0 / 29.0)) * (3.0 * delta ** 2.0)) # multiply in white values xyz_array[:, 0] *= wht_x xyz_array[:, 1] *= wht_y xyz_array[:, 2] *= wht_z # convert to sRGB using the specified conversion matrix rgb_out = numpy.asarray(numpy.dot(xyz_array, conversionMatrix.T)) # apply sRGB gamma correction if requested if transferFunc is not None: rgb_out = transferFunc(rgb_out, **kwargs) # clip unrepresentable colors if requested if clip: rgb_out = numpy.clip(rgb_out, 0.0, 1.0) # make the output match the dimensions/shape of input if orig_dim == 1: rgb_out = rgb_out[0] elif orig_dim == 3: rgb_out = numpy.reshape(rgb_out, orig_shape) return rescaleColor(rgb_out, convertTo='psychopy')
[docs]def cielch2rgb(lch, whiteXYZ=None, conversionMatrix=None, transferFunc=None, clip=False, **kwargs): """Transform CIE L*C*h* coordinates to RGB tristimulus values. Parameters ---------- lch : tuple, list or ndarray 1-, 2-, 3-D vector of CIE L*C*h* coordinates to convert. The last dimension should be length-3 in all cases specifying a single coordinate. The hue angle *h is expected in degrees. whiteXYZ : tuple, list or ndarray 1-D vector coordinate of the white point in CIE-XYZ color space. Must be the same white point needed by the conversion matrix. The default white point is D65 if None is specified, defined as X, Y, Z = 0.9505, 1.0000, 1.0890 conversionMatrix : tuple, list or ndarray 3x3 conversion matrix to transform CIE-XYZ to RGB values. The default matrix is sRGB with a D65 white point if None is specified. Note that values must be gamma corrected to appear correctly according to the sRGB standard. transferFunc : pyfunc or None Signature of the transfer function to use. If None, values are kept as linear RGB (it's assumed your display is gamma corrected via the hardware CLUT). The TF must be appropriate for the conversion matrix supplied. Additional arguments to 'transferFunc' can be passed by specifying them as keyword arguments. Gamma functions that come with PsychoPy are 'srgbTF' and 'rec709TF', see their docs for more information. clip : boolean Make all output values representable by the display. However, colors outside of the display's gamut may not be valid! Returns ------- ndarray array of RGB tristimulus values """ lch, orig_shape, orig_dim = unpackColors(lch) # convert values to L*a*b* lab = numpy.empty(lch.shape, dtype=lch.dtype) lab[:, 0] = lch[:, 0] lab[:, 1] = lch[:, 1] * numpy.math.cos(numpy.math.radians(lch[:, 2])) lab[:, 2] = lch[:, 1] * numpy.math.sin(numpy.math.radians(lch[:, 2])) # convert to RGB using the CIE L*a*b* function rgb_out = cielab2rgb(lab, whiteXYZ=whiteXYZ, conversionMatrix=conversionMatrix, transferFunc=transferFunc, clip=clip, **kwargs) # make the output match the dimensions/shape of input if orig_dim == 1: rgb_out = rgb_out[0] elif orig_dim == 3: rgb_out = numpy.reshape(rgb_out, orig_shape) return rgb_out # don't do signed RGB conversion, done by cielab2rgb
[docs]def dkl2rgb(dkl, conversionMatrix=None): """Convert from DKL color space (Derrington, Krauskopf & Lennie) to RGB. Requires a conversion matrix, which will be generated from generic Sony Trinitron phosphors if not supplied (note that this will not be an accurate representation of the color space unless you supply a conversion matrix). Examples -------- Converting a single DKL color to RGB:: dkl = [90, 0, 1] rgb = dkl2rgb(dkl, conversionMatrix) """ # make sure the input is an array dkl = numpy.asarray(dkl) if conversionMatrix is None: conversionMatrix = numpy.asarray([ # (note that dkl has to be in cartesian coords first!) # LUMIN %L-M %L+M-S [1.0000, 1.0000, -0.1462], # R [1.0000, -0.3900, 0.2094], # G [1.0000, 0.0180, -1.0000]]) # B logging.warning('This monitor has not been color-calibrated. ' 'Using default DKL conversion matrix.') if len(dkl.shape) == 3: dkl_NxNx3 = dkl # convert a 2D (image) of Spherical DKL colours to RGB space origShape = dkl_NxNx3.shape # remember for later NxN = origShape[0] * origShape[1] # find nPixels dkl = numpy.reshape(dkl_NxNx3, [NxN, 3]) # make Nx3 rgb = dkl2rgb(dkl, conversionMatrix) # convert return numpy.reshape(rgb, origShape) # reshape and return else: dkl_Nx3 = dkl # its easier to use in the other orientation! dkl_3xN = numpy.transpose(dkl_Nx3) if numpy.size(dkl_3xN) == 3: RG, BY, LUM = sph2cart(dkl_3xN[0], dkl_3xN[1], dkl_3xN[2]) else: RG, BY, LUM = sph2cart(dkl_3xN[0, :], dkl_3xN[1, :], dkl_3xN[2, :]) dkl_cartesian = numpy.asarray([LUM, RG, BY]) rgb = numpy.dot(conversionMatrix, dkl_cartesian) # return in the shape we received it: return numpy.transpose(rgb)
[docs]def dklCart2rgb(LUM, LM, S, conversionMatrix=None): """Like dkl2rgb except that it uses cartesian coords (LM,S,LUM) rather than spherical coords for DKL (elev, azim, contr). NB: this may return rgb values >1 or <-1 """ NxNx3 = list(LUM.shape) NxNx3.append(3) dkl_cartesian = numpy.asarray( [LUM.reshape([-1]), LM.reshape([-1]), S.reshape([-1])]) if conversionMatrix is None: conversionMatrix = numpy.asarray([ # (note that dkl has to be in cartesian coords first!) # LUMIN %L-M %L+M-S [1.0000, 1.0000, -0.1462], # R [1.0000, -0.3900, 0.2094], # G [1.0000, 0.0180, -1.0000]]) # B rgb = numpy.dot(conversionMatrix, dkl_cartesian) return numpy.reshape(numpy.transpose(rgb), NxNx3)
[docs]def rgb2hsv(rgb): """Convert values from linear RGB to HSV colorspace. Parameters ---------- rgb : `array_like` 1-, 2-, 3-D vector of RGB coordinates to convert. The last dimension should be length-3 in all cases, specifying a single coordinate. Returns ------- ndarray HSV values with the same shape as the input. """ # Based on https://www.geeksforgeeks.org/program-change-rgb-color-model-hsv-color-model/ rgb, orig_shape, orig_dim = unpackColors(rgb) # need to rescale RGB values to 0.0 and 1.0 rgb = rescaleColor(rgb, convertTo='unsigned') # get row min/max indices rmax = numpy.argmax(rgb, axis=1) rmin = numpy.argmin(rgb, axis=1) # get min/max values for each color coordinate sel = numpy.arange(len(rgb)) cmax = rgb[sel, rmax] cmin = rgb[sel, rmin] # compute the difference between the max and min color value delta = cmax - cmin # vector to return HSV values hsv_out = numpy.zeros_like(rgb, dtype=float) # --- calculate vibrancy --- dzero = delta == 0 # if delta is zero the color is a shade of grey inv_dzero = None if numpy.any(dzero): # vibrancy is 1 hsv_out[dzero, 2] = numpy.sum(rgb[dzero], axis=1) / 3. inv_dzero = ~dzero if inv_dzero is not None: hsv_out[inv_dzero, 2] = cmax[inv_dzero] else: hsv_out[:, 2] = cmax[:] # no B/W colors # --- calculate saturation --- hsv_out[:, 1] = numpy.where(cmax > 0.0, delta / cmax, 0.0) # --- calculate hues --- # views of each column r = rgb[:, 0] g = rgb[:, 1] b = rgb[:, 2] # select on rows where the RGB gun value is max and not `dzero` sel_r = (rmax == 0) & inv_dzero if inv_dzero is not None else rmax == 0 sel_g = (rmax == 1) & inv_dzero if inv_dzero is not None else rmax == 1 sel_b = (rmax == 2) & inv_dzero if inv_dzero is not None else rmax == 2 if numpy.any(sel_r): # if red == cmax hsv_out[sel_r, 0] = \ (60 * ((g[sel_r] - b[sel_r]) / delta[sel_r]) + 360) % 360 if numpy.any(sel_g): # if green == cmax hsv_out[sel_g, 0] = \ (60 * ((b[sel_g] - r[sel_g]) / delta[sel_g]) + 120) % 360 if numpy.any(sel_b): # if blue == cmax hsv_out[sel_b, 0] = \ (60 * ((r[sel_b] - g[sel_b]) / delta[sel_b]) + 240) % 360 # round the hue angle value hsv_out[:, 0] = numpy.round(hsv_out[:, 0]) # make the output match the dimensions/shape of input if orig_dim == 1: hsv_out = hsv_out[0] elif orig_dim == 3: hsv_out = numpy.reshape(hsv_out, orig_shape) return hsv_out
[docs]def hsv2rgb(hsv_Nx3): """Convert from HSV color space to RGB gun values. usage:: rgb_Nx3 = hsv2rgb(hsv_Nx3) Note that in some uses of HSV space the Hue component is given in radians or cycles (range 0:1]). In this version H is given in degrees (0:360). Also note that the RGB output ranges -1:1, in keeping with other PsychoPy functions. """ # based on method in # http://en.wikipedia.org/wiki/HSL_and_HSV#Converting_to_RGB hsv_Nx3 = numpy.asarray(hsv_Nx3, dtype=float) # we expect a 2D array so convert there if needed origShape = hsv_Nx3.shape hsv_Nx3 = hsv_Nx3.reshape([-1, 3]) H_ = old_div((hsv_Nx3[:, 0] % 360), 60.0) # this is H' in the wikipedia version # multiply H and V to give chroma (color intensity) C = hsv_Nx3[:, 1] * hsv_Nx3[:, 2] X = C * (1 - abs(H_ % 2 - 1)) # rgb starts rgb = hsv_Nx3 * 0 # only need to change things that are no longer zero II = (0 <= H_) * (H_ < 1) rgb[II, 0] = C[II] rgb[II, 1] = X[II] II = (1 <= H_) * (H_ < 2) rgb[II, 0] = X[II] rgb[II, 1] = C[II] II = (2 <= H_) * (H_ < 3) rgb[II, 1] = C[II] rgb[II, 2] = X[II] II = (3 <= H_) * (H_ < 4) rgb[II, 1] = X[II] rgb[II, 2] = C[II] II = (4 <= H_) * (H_ < 5) rgb[II, 0] = X[II] rgb[II, 2] = C[II] II = (5 <= H_) * (H_ < 6) rgb[II, 0] = C[II] rgb[II, 2] = X[II] m = (hsv_Nx3[:, 2] - C) rgb += m.reshape([len(m), 1]) # V-C is sometimes called m return rgb.reshape(origShape) * 2 - 1
[docs]def lms2rgb(lms_Nx3, conversionMatrix=None): """Convert from cone space (Long, Medium, Short) to RGB. Requires a conversion matrix, which will be generated from generic Sony Trinitron phosphors if not supplied (note that you will not get an accurate representation of the color space unless you supply a conversion matrix) usage:: rgb_Nx3 = lms2rgb(dkl_Nx3(el,az,radius), conversionMatrix) """ # its easier to use in the other orientation! lms_3xN = numpy.transpose(lms_Nx3) if conversionMatrix is None: cones_to_rgb = numpy.asarray([ # L M S [4.97068857, -4.14354132, 0.17285275], # R [-0.90913894, 2.15671326, -0.24757432], # G [-0.03976551, -0.14253782, 1.18230333]]) # B logging.warning('This monitor has not been color-calibrated. ' 'Using default LMS conversion matrix.') else: cones_to_rgb = conversionMatrix rgb = numpy.dot(cones_to_rgb, lms_3xN) return numpy.transpose(rgb) # return in the shape we received it
[docs]def rgb2lms(rgb_Nx3, conversionMatrix=None): """Convert from RGB to cone space (LMS). Requires a conversion matrix, which will be generated from generic Sony Trinitron phosphors if not supplied (note that you will not get an accurate representation of the color space unless you supply a conversion matrix) usage:: lms_Nx3 = rgb2lms(rgb_Nx3(el,az,radius), conversionMatrix) """ # its easier to use in the other orientation! rgb_3xN = numpy.transpose(rgb_Nx3) if conversionMatrix is None: cones_to_rgb = numpy.asarray([ # L M S [4.97068857, -4.14354132, 0.17285275], # R [-0.90913894, 2.15671326, -0.24757432], # G [-0.03976551, -0.14253782, 1.18230333]]) # B logging.warning('This monitor has not been color-calibrated. ' 'Using default LMS conversion matrix.') else: cones_to_rgb = conversionMatrix rgb_to_cones = numpy.linalg.inv(cones_to_rgb) lms = numpy.dot(rgb_to_cones, rgb_3xN) return numpy.transpose(lms) # return in the shape we received it
[docs]def rgb2dklCart(picture, conversionMatrix=None): """Convert an RGB image into Cartesian DKL space. """ # Turn the picture into an array so we can do maths picture = numpy.array(picture) # Find the original dimensions of the picture origShape = picture.shape # this is the inversion of the dkl2rgb conversion matrix if conversionMatrix is None: conversionMatrix = numpy.asarray([ # LUMIN-> %L-M-> L+M-S [0.25145542, 0.64933633, 0.09920825], [0.78737943, -0.55586618, -0.23151325], [0.26562825, 0.63933074, -0.90495899]]) logging.warning('This monitor has not been color-calibrated. ' 'Using default DKL conversion matrix.') else: conversionMatrix = numpy.linalg.inv(conversionMatrix) # Reshape the picture so that it can multiplied by the conversion matrix red = picture[:, :, 0] green = picture[:, :, 1] blue = picture[:, :, 2] dkl = numpy.asarray([red.reshape([-1]), green.reshape([-1]), blue.reshape([-1])]) # Multiply the picture by the conversion matrix dkl = numpy.dot(conversionMatrix, dkl) # Reshape the picture so that it's back to it's original shape dklPicture = numpy.reshape(numpy.transpose(dkl), origShape) return dklPicture

Back to top