how does the function "color correction" work? where to buy an color card ?

[bobshipwood]

refer to https://plantcv.org/tutorials/simple-rgb-workflow :

1 i need buy an color card first, but which type of color card should i bought？
"ColorChecker" or “SpyderCHECKR”?

2 then i need to take a picture with a plant and the color card, but the color card place where?
it`s left side on the plant in https://plantcv.org/tutorials/simple-rgb-workflow,

is it true that whether i place the color card in the left side or right side, the program cat verfy it`s place (pcv.visualize.colorspaces(rgb_img=crop_img))

[bobshipwood]

and also in https://plantcv.org/tutorials/color-correction
“

Detect the color card, create a matrix of color chip values, and perform linear color correction.

img_cc = pcv.transform.auto_correct_color(rgb_img=crop_img, radius=15)

”

what does the function actitualy work indeed？

[nfahlgren]

We have links to color cards we have tested on the documentation page for pcv.transform.detect_color_card: https://plantcv.readthedocs.io/en/latest/transform_detect_color_card/. But SpyderCHECKR should also work because we should be able to support any Macbeth-style color card.

You can place the color card anywhere in the image. PlantCV can detect it automatically using the pcv.transform.auto_correct_color or pcv.transform.detect_color_card functions. But we recommend that the card be placed consistently in one place, away from the plant or other objects of interest, and preferably flat and square with the image (though we can detect the card even if it is rotated).

pcv.transform.auto_correct_color runs a few other PlantCV functions to make color correction easier. It detects the color card by segmenting the image using an auto-thresholding method then searches the image for approximately square contours above a size threshold. If it successfully detects color squares within the color card it estimates the chip grid layout and creates a color card mask (1 spot per chip) and extracts the observed RGB values. It then builds an affine transformation matrix that transforms the observed RGB values toward the standard RGB values for the Macbeth style color checker cards.

[bobshipwood]

below is my code and error，my question is that what is the functions parameter radius mean？ in my case，i fill radius=15 into the function，but i receive an runtimes error "RuntimeError: No color card found"

code:

from plantcv import plantcv as pcv

image = "./color.png"
img, path, filename = pcv.readimage(filename=image)

#crop_img = pcv.crop(img=img, x=240, y=100, w=280, h=250)
# print(type(crop_img))
#pcv.plot_image(crop_img)

# Detect the color card, create a matrix of color chip values, and perform linear color correction.
img_cc = pcv.transform.auto_correct_color(rgb_img=img, radius=15)
print(type(img_cc))

error:
D:\yolo.venv\Scripts\python.exe D:\yolo\10.pv.py
DeprecationWarning: The 'label' parameter is no longer utilized, since color chip size is now metadata. It will be removed in PlantCV v5.0. Current PlantCV version: 4.9
DeprecationWarning: The 'label' parameter is no longer utilized, since color chip size is now metadata. It will be removed in PlantCV v5.0. Current PlantCV version: 4.9
D:\yolo.venv\lib\site-packages\numpy_core\fromnumeric.py:3596: RuntimeWarning: Mean of empty slice.
return _methods._mean(a, axis=axis, dtype=dtype,
D:\yolo.venv\lib\site-packages\numpy_core_methods.py:138: RuntimeWarning: invalid value encountered in scalar divide
ret = ret.dtype.type(ret / rcount)
Traceback (most recent call last):
File "D:\yolo\10.pv.py", line 11, in
img_cc = pcv.transform.auto_correct_color(rgb_img=img, radius=15)
File "D:\yolo.venv\lib\site-packages\plantcv\plantcv\transform\auto_correct_color.py", line 42, in auto_correct_color
labeled_mask = detect_color_card(rgb_img=rgb_img, color_chip_size=color_chip_size, roi=roi, **kwargs)
File "D:\yolo.venv\lib\site-packages\plantcv\plantcv\transform\detect_color_card.py", line 387, in detect_color_card
sub_mask, debug_img, marea, mheight, mwidth, _ = _rect_filter(rgb_img,
File "D:\yolo.venv\lib\site-packages\plantcv\plantcv_helpers.py", line 843, in _rect_filter
return function(sub_img, **kwargs)
File "D:\yolo.venv\lib\site-packages\plantcv\plantcv\transform\detect_color_card.py", line 207, in _color_card_detection
fatal_error('No color card found')
File "D:\yolo.venv\lib\site-packages\plantcv\plantcv\fatal_error.py", line 13, in fatal_error
raise RuntimeError(error)
RuntimeError: No color card found

[HaleySchuhl]

Hi @bobshipwood ,

The radius parameter specifies the radius of circles to made in the color card's labeled mask (default = 20).

There are other parameters one can tune, documented here https://plantcv.readthedocs.io/en/latest/transform_detect_color_card/. If you'd like, you can upload the image you are running and we can try to tune the detection settings.

[bobshipwood]

@HaleySchuhl,

j just not undestand what "radius" stand for? what affect when i just trun it higher or turn it lower，

below is my picture when i input （color.jpg）， and output form is color.png, did i make it right? i doubt i have miss some important step,

and how could i saw the last photo with 24 little blue point in https://plantcv.org/tutorials/color-correction?

image = "./color.jpg"
img, path, filename = pcv.readimage(filename=image)

#crop_img = pcv.crop(img=img, x=240, y=100, w=280, h=250)
# print(type(crop_img))
#pcv.plot_image(crop_img)

# Detect the color card, create a matrix of color chip values, and perform linear color correction.
img_cc = pcv.transform.auto_correct_color(rgb_img=img, radius=15)
print(pcv.outputs.metadata['median_color_chip_size']['value'])

pcv.print_image(img=img_cc, filename="./color.png")

[HaleySchuhl]

The radius parameter changes the radius of regions that get sampled for creating the color matrix. Looks like you are using the function correctly, just would need to turn pcv.params.debug = "plot" to visualize the debug image, which aims to show what is being done at each step.

from plantcv import plantcv as pcv 

image = "/Users/hschuhl/Desktop/color.jpg"
pcv.params.debug = "plot" 
img, path, filename = pcv.readimage(filename=image)

#crop_img = pcv.crop(img=img, x=240, y=100, w=280, h=250)
# print(type(crop_img))
#pcv.plot_image(crop_img)
_ = pcv.transform.detect_color_card(img, radius=15)
_ = pcv.transform.detect_color_card(img, radius=5)

# Detect the color card, create a matrix of color chip values, and perform linear color correction.
img_cc = pcv.transform.auto_correct_color(rgb_img=img, radius=15)
print(pcv.outputs.metadata['median_color_chip_size']['value'])

pcv.print_image(img=img_cc, filename="./corrected_color.png")

[bobshipwood]

@HaleySchuhl
thanks a lot!! by the way,if i turn the radius to 1, just like your second last picture， what will effect in color correction?
it seem`s the same in radius=15 and radius =1?

[bobshipwood]

and by the way again,, i take the photo with white color`s Background, it seems the white color turns a little green in your last picture.

why?

[kmurphy61]

@bobshipwood you can see suggestions for how to place a color card in this publication. https://app.jove.com/t/67619/imaging-analysis-for-quantifying-maize-zea-mays-abiotic-stress

Your images look oversaturated, you need to adjust your camera settings and can see details on how to do so in the publication linked here. Color correction cannot fix images that are too over or undersaturated, this is a problem with image capture.

[HaleySchuhl]

I can confirm there's quite a bit of the photo that is oversaturated. We have a function (https://plantcv.readthedocs.io/en/latest/saturation_threshold/) to check for this, where the function outputs a mask where oversaturated regions are black.

sat_thresh = pcv.threshold.saturation(rgb_img=img, threshold=255, channel="any")

The radius parameter only changes the radius of region of an image that get sampled, and this hue data is used to create the color matrix. A smaller radius samples a smaller region, so likely shouldn't impact the result of color correction provided the mask alignment is accurate.

[bobshipwood]

@HaleySchuhl > where the function outputs a mask where oversaturated regions are black.

ths a lot to you!!

[bobshipwood]

i have question again，i changed my photo`s situtation with my background in black。

after function test:
sat_thresh = pcv.threshold.saturation(rgb_img=img, threshold=255, channel="any") :

1 it seems a little black point in the cap of the bottle，dose it matter in color correction？ or i must clear all the black point。

2 the last picture i post is about color card. it seems 4 of the 24 color grid does not surrounded by a pink circle？ it seems have am impact on color correction or not?

[HaleySchuhl]

Is there a difference in the color correction result if you mask out the oversaturated pixels? I would expect it to only impact color correction if the color card is affected by over-saturation.

[bobshipwood]

only 0.1% black point(after over-saturation test) in my picture, so do i need to clean all the black point ?

[HaleySchuhl]

Are you trying to measure a color phenotype in this dataset? I see no evidence that such a small proportion of over-saturated pixels to impact color correction. I tried to mask out the region and do not see any impact on color correction. It does appear that your images are possibly too dark now.

%matplotlib widget

from plantcv import plantcv as pcv 

image = "/Users/hschuhl/Desktop/jar.jpg"

pcv.params.debug = "plot"
pcv.params.verbose = False
img, path, filename = pcv.readimage(filename=image)


# Detect the color card, create a matrix of color chip values, and perform linear color correction.
img_cc = pcv.transform.auto_correct_color(rgb_img=img, radius=25)

## Check the color correction plot to see how much the colorspace is changing. 
mask = pcv.transform.detect_color_card(img, radius=15)
color_matrix = pcv.transform.get_color_matrix(rgb_img=img, mask=mask)
img_cc = pcv.transform.auto_correct_color(rgb_img=img, radius=25)
corrected_matrix = pcv.transform.get_color_matrix(rgb_img=img_cc, mask=mask)
std_matrix = pcv.transform.std_color_matrix(pos=3)

_ = pcv.visualize.color_correction_plot(color_matrix, std_matrix, corrected_matrix)

Compare to result after removing over-saturated pixels:

sat_thresh = pcv.threshold.saturation(rgb_img=img, threshold=252, channel="any")
masked_rgb = pcv.apply_mask(img=img, mask=sat_thresh, mask_color="black")
img_cc = pcv.transform.auto_correct_color(rgb_img=masked_rgb, radius=25)

# Check the plot for a difference 
mask = pcv.transform.detect_color_card(img, radius=15)
color_matrix = pcv.transform.get_color_matrix(rgb_img=img, mask=mask)
img_cc = pcv.transform.auto_correct_color(rgb_img=img, radius=25)
corrected_matrix = pcv.transform.get_color_matrix(rgb_img=img_cc, mask=mask)
std_matrix = pcv.transform.std_color_matrix(pos=3)

_ = pcv.visualize.color_correction_plot(color_matrix, std_matrix, corrected_matrix)

• Note, I'm using an un-released function pcv.visualize.color_correction_plot to compare the observed color values to the expected values. This feature is slotted for plantcv v5.0 currently. Color chip comparison #1779