How to perform color correction?

[hardikdava]

Question

I want to performs a color correction. I am looking for an example script but could not find it. I have used this color checker and I have video as source data. How can I get final color corrected frames/video?

Thanks in advance.

[KelSolaar]

Hi @hardikdava,

Do you already have the measured values or do you need to sample them first?

[hardikdava]

@KelSolaar Thanks for your reply. I have captured images like this.

[KelSolaar]

Is it possible to share your entire function, it is pretty much impossible to guess what is going on here.

[hardikdava]

I have sequence of frames which is inside directory data/input. Some initial frames has color checker. if color checker is found then computer color checker matrix and apply it to all the frames.

from pathlib import Path
import colour
import supervision as sv

import matplotlib
matplotlib.use('TkAgg')

from colour_checker_detection import (
    detect_colour_checkers_inference)

colour.plotting.colour_style()

COLOUR_CHECKER_IMAGE_PATHS = sv.list_files_with_extensions("data/input")

COLOUR_CHECKER_IMAGES = [
    colour.cctf_decoding(colour.io.read_image(path))
    for path in COLOUR_CHECKER_IMAGE_PATHS
]

SWATCHES = []
for image in COLOUR_CHECKER_IMAGES:
    if len(SWATCHES) > 0:
        break
    for colour_checker_data in detect_colour_checkers_inference(
            image, additional_data=False):
        SWATCHES.append(colour_checker_data)

# D65 = colour.CCS_ILLUMINANTS['CIE 1931 2 Degree Standard Observer']['D65']
REFERENCE_COLOUR_CHECKER = colour.CCS_COLOURCHECKERS[
    'ColorChecker24 - After November 2014']


REFERENCE_SWATCHES = colour.XYZ_to_RGB(
    colour.xyY_to_XYZ(list(REFERENCE_COLOUR_CHECKER.data.values())),
    'sRGB', REFERENCE_COLOUR_CHECKER.illuminant)

input_images = COLOUR_CHECKER_IMAGE_PATHS
output_dir = Path("data/cc_corrected")
for i, image_path in enumerate(input_images):

    image = colour.cctf_decoding(colour.io.read_image(image_path))
    img = colour.colour_correction(
            image, SWATCHES[0], REFERENCE_SWATCHES, method="Finlayson 2015")

    output_path = output_dir / image_path.name

    img = colour.cctf_encoding(img)
    colour.write_image(img, output_path, bit_depth="uint8")

[KelSolaar]

Hi,

It is an integer overflow, we never ever clamp anything in colour, even when writing an image! What is happening is that your image has now negative values because of the colour correction matrix and you then try to write those into a 8-bit container. You could do this when writing your image:

colour.write_image(np.clip(img, 0, 1), output_path, bit_depth="uint8")

PS: The reason we do not clamp is to that one can see that there is potentially data loss.

[hardikdava]

@KelSolaar Understood. Thanks for your help. This is the working version, if anyone wants to use it.

from pathlib import Path
import colour
import supervision as sv
import numpy as np

"""
pip install colour-science colour-checker-detection numpy supervision
"""

from colour_checker_detection import (
    detect_colour_checkers_inference)

colour.plotting.colour_style()

input_dir = Path("data/input")
output_dir = Path("data/cc_corrected")
COLOUR_CHECKER_IMAGE_PATHS = sv.list_files_with_extensions(input_dir)

COLOUR_CHECKER_IMAGES = [
    colour.cctf_decoding(colour.io.read_image(path))
    for path in COLOUR_CHECKER_IMAGE_PATHS
]

SWATCHES = []

for image in COLOUR_CHECKER_IMAGES:
    if len(SWATCHES) > 0:
        break
    for colour_checker_data in detect_colour_checkers_inference(
            image, additional_data=False):
        SWATCHES.append(colour_checker_data)

# D65 = colour.CCS_ILLUMINANTS['CIE 1931 2 Degree Standard Observer']['D65']
REFERENCE_COLOUR_CHECKER = colour.CCS_COLOURCHECKERS[
    'ColorChecker24 - After November 2014']

REFERENCE_SWATCHES = colour.XYZ_to_RGB(
    colour.xyY_to_XYZ(list(REFERENCE_COLOUR_CHECKER.data.values())),
    'sRGB', REFERENCE_COLOUR_CHECKER.illuminant)

input_images = COLOUR_CHECKER_IMAGE_PATHS

for i, image_path in enumerate(input_images):

    image = colour.cctf_decoding(colour.io.read_image(image_path))
    img = colour.colour_correction(
            image, SWATCHES[0], REFERENCE_SWATCHES, method="Finlayson 2015")

    output_path = output_dir / image_path.name

    img = colour.cctf_encoding(img)
    colour.write_image(np.clip(img, 0, 1), output_path, bit_depth="uint8")