Biomedical Imaging Research Unit

School of Medical Sciences

Faculty of Medical and Health Sciences
The University of Auckland
Private Bag 92019
Auckland, NZ
Ph: 373 7599 ext. 87438
http://www.health.auckland.ac.nz/biru/
[email protected]

.
Colour Analysis Tools in ImageJ 11 July 2007
Jacqui Ross

Digital Images
= arrays of pixels (picture elements). Each has a value and is made up of bits (8bit, 12bit, 16bit)

Colour Modes/Colour Spaces (3D)

http://en.wikipedia.org/wiki/Color_space

RGB
= Red Green Blue
Starts with black and as colours are added, the amount of emitted light is increased (i.e. additive
colour) = Red, Green and Blue « channels » with pixel values for every pixel in each channel
= colour digital cameras, computer display, etc.
For example, for a 24bit colour image, white = R255, G255, B255; black = R0, G0, B0.
Pure red = R255, G0, B0, green = R0, G255, B0, blue = R0, G0, B255.
All other colours are displayed by varying these pixel values.

HSB/HSV
= Hue Saturation Brightness/Value
Each colour shade (hue) has a value.
Saturation is equivalent to intensity, Brightness/Value relate to lightness.
Work in combination. Relates to RGB values.

CIE Lab
= Luminance (L) and two colour components (a, b) which work in an opposing way.
Attempts to approximate human vision, therefore has a larger colour gamut.
L* = 0 = black
L* =100 = white
YUV
= Luma or brightness (Y) + colour signal (U, V combined)
Used for analog TV for creating the picture, e.g. NTSC, PAL, SECAM formats

CMY(K)
= Cyan Magenta Yellow
Starts with white and as colours are added, the amount of absorbed light is increased (i.e.
subtractive colour).
= for pigments, printing, etc.
M + Y = red
C + Y = green
M + C = blue
K (black) is added to improve black since C + M + Y is not true black.

Colour Segmentation

RGB Channel Separation

Fluorescence/confocal images
Split into individual channels.
Image – Color – RGB Split. This will give you 3 separate grayscale images which you can then
work on independently.

Histological stains (e.g. DAB, AEC, etc.)

First try splitting channels as for fluorescence images since this is the simplest way to work with the
images (Image – Color – RGB split). Usually doesn’t work because stains are not “pure” colours.

Specialised Plugins and Tools

• Color Histogram
• RGB Recolor
• RGB Measure
• RGB Measure Plus
• Color Profiler
• Colour Space Converter
• Convert RGB to Luv, Lab or HSV
• Threshold Colour

2
 • Colour Deconvolution
• 3D Color Inspector

Colour Histogram
http://rsb.info.nih.gov/ij/plugins/color-histogram.html
Gives you histograms of R, G, B values and mean pixel values in a Results table. Works on the
whole image and ROIs.

RGB Recolor
http://rsb.info.nih.gov/ij/plugins/recolor.html
Allows linear alteration of the colors in R, G and B channels of RGB images, where:
New color = Old Color*Scaling Factor + Constant
If the new color is greater than 255 it is set to 255. If it is less than 0 it is set to 0.

3
RGB Measure
http://rsb.info.nih.gov/ij/plugins/rgb-measure.html
Gives you measurements of R, G, B values. Also gives you any other measurements (in R, G, B
channels) as selected in Analyze – Set Measurements, e.g. area, median, min, max, etc. Works on
the whole image and ROIs.
Plugins – RGB Measure

RGB Measure Plus

http://rsb.info.nih.gov/ij/plugins/rgb-measure-plus.html
Same as RGB Measure except that you can define thresholds.

4
Color Profiler
http://rsb.info.nih.gov/ij/plugins/color-profiler.html
A useful tool to determine proportions of R, G, B in the image. Provides a plot of intensity for each
channel as well as tabulated results.

Color Space Converter

http://rsb.info.nih.gov/ij/plugins/color-space-converter.html

Convert RGB to Luv, Lab or HSV

http://rsb.info.nih.gov/ij/plugins/color-converters.html

5
Threshold Colour
http://imagejdocu.tudor.lu/imagej-documentation-wiki/plugins/threshold_colour

This plugin is designed as a filter to either pass or stop parts of the image being displayed
depending on the settings you have selected.
You can choose to see the original image or alternatively select threshold to see your selection in
black on white.
There are 4 colour modes available; HSB, RGB, CIE Lab and YUV.
If you find it useful, you can convert your image using additional plugins such as the
Color Space Converter
HSB
Example 1: Haematoxylin & Eosin stained skin section

You can see a number of peaks present in the Hue histogram. The ones to the right represent the
purple/pink colours present in the image. The green one represents the background which is not
totally white.

6
Change Pass to Stop to see the other colour (purple). This allows you to check the segmentation.
You can also change between Original and Filtered.

7
Example 2: DAB/Nissl stained brain section
HSB is a good colour mode to use for DAB because it is often quite dark (saturated) in comparison
with the counterstain used (e.g. Nissl, Haematoxylin) so it’s easier to segment out. In addition,
although DAB has a strong blue component, the hue is quite different from the counterstain.

You can use the Colour Space Converter first to see whether HSB shows the components you
want to separate as distinct.

You can then choose to use the original image for segmentation or the newly converted image.

8
Look at the histograms. If there are a number of peaks present in the hue histogram, it will be
easier to segment the image.

9
Segmentation of Nissl staining

Blocks hues 0-160

Passes hues 160-255

Passes all

Passes all

10
Click on Threshold to get a black/white representation. This is still an RGB image. If you want to
keep it for analysis, RHS mouse click and Duplicate, Rename and save. This is the segmented
Nissl image.

You can binarise the segmented image by going to Image – Process – Make Binary.

11
You can then create masks/selections for the ROI Manager by going to the Edit menu, e.g. Edit –
Select – Create Selection and then Edit – Selection – Add to Manager. The ROI Manager will then
open and add the selection as a ROI.

Click on Original and turn off Threshold to begin segmentation of the DAB staining.

12
Segmentation of DAB staining
You could start by changing Pass to Stop (Hue histogram). This will stop the Nissl staining coming
through. All other hues will pass which means we now see the DAB staining.

Passes hues 0-160

Blocks hues 160-255

Adjust the Saturation and Brightness sliders until only the DAB staining shows through.

13
Click on Threshold.

14
RHS mouse click, Duplicate, Rename, Binarise (Process – Make Binary) and save.

Note that for measuring areas/particles, etc. of binarised images, it is the black area (on white) that
is measured as the object of interest.

15
16
Colour Deconvolution
http://imagejdocu.tudor.lu/imagej-documentation-wiki/plugins/colour_deconvolution

http://www.dentistry.bham.ac.uk/landinig/software/cdeconv/cdeconv.html

Theory: http://home.planet.nl/~ber03728/4N6site/improc/decoplugin/decoexpl/p01.htm

Histological stains are “light absorbing dyes” so can be considered as being subtractive colour.
Requires neutral background to work properly. Vectors need to be worked out from single-stained
control slides or from ROIs where you can be confident that only one stain is present. The vectors
are determined from the optical density of these areas and then normalised.

Vectors have been developed for the following stain combinations:

• Haematoxylin and Eosin (H&E)

• Haematoxylin and DAB (H DAB)
• Fast Red, Fast Blue and DAB
• Methyl green and DAB
• Haematoxylin, Eosin and DAB (H&E DAB)
• Haematoxylin and AEC (H AEC)
• Azan-Mallory
• Alcian blue & Haematoxylin
• Haematoxylin and Periodic Acid of Schiff (PAS)
• RGB subtractive
• CMY subtractive
• User values entered by hand
• Values interactively determined from rectangular ROIs

Example Haematoxylin & Eosin

Analyze – Tools – Color Histogram

First check that the background is neutral. This is best done using a Region of Interest in an
unstained area.
Draw a ROI in a “white” space. Then go to Analyze – Tools – Color Histogram. The R, G, B values
should be similar. If they aren’t, then you need to adjust the image using background subtraction
(Process –subtract background), Image – Adjust processes, or by subtracting a background
image using Process – Image Calculator.
Go to Plugins – Colour_analysis-Color Deconvolution (or wherever you have loaded the
plugin).

17
Using built-in vectors
Select the vector you want which most closely resembles your image (e.g. H & E), show matrices
and click OK.
If there are only 2 colours present that you want to separate, (as in this case Haematoxylin from
Eosin), then the result should be one image of each corresponding to the individual stain and
another one which is white. You also get a log file and the matrix data. The values have been
determined experimentally by the lab group of the person who has developed this plugin.

Using the ROI to define your own vectors

If you have a different stain from those available in the plugin, you can determine your own matrix if
you have some clear areas where only one stain is present (using ROI). Where two stains coexist,
it may be difficult to separate them out and this “mixed” region may show up in the 3rd image rather
than it being white. In this case, you can develop your own vectors experimentally.

18
Developing your own vectors experimentally
Create one control slide for each individual stain. You then determine the vectors for each stain
using the ROI. You can then combine the vectors to get the matrix to apply to the dual/triple stained
slides.

Note that this plugin can be used with stacks.

3D Color Inspector
http://rsb.info.nih.gov/ij/plugins/color-inspector.html
http://www.f4.fhtw-berlin.de/~barthel/ImageJ/ColorInspector//help.htm
Allows you to change the colour display and save a modified image.

By changing the colour rotation (or other attributes), you can improve contrast which may make
segmentation easier.

19
You can also do simple colour segmentation within 3D Inspector as shown below:

Or segment the modified image using other plugins, e.g. Threhold Colour.

20
21
Image Correction Tools

Built in Command: Process – Subtract Background

http://rsb.info.nih.gov/ij/docs/menus/process.html
Uses rolling ball algorithm, choose radius according to the size of features, choose light
background for brightfield colour images.

Calculator Plus
http://rsb.info.nih.gov/ij/plugins/calculator-plus.html
Uses image and blank image together with a multiplication factor of 255 as shown below:

Shading Corrector
http://rsb.info.nih.gov/ij/plugins/shading-corrector.html
Works only on 8 bit images. Requires a blank image also as for Calculator Plus.

Colour_Correct
Available at: http://www.dentistry.bham.ac.uk/landinig/software/software.html

22
Does not correct for uneven illumination but compensates for inadequate lamp temperature by
allowing the setting of “black” and “white” points.

Background Correction
Grayscale fluorescence 8bit images only at the moment. Corrects for uneven illumination.
http://rsb.info.nih.gov/ij/plugins/background.html

More….
If you want to read more about image correction for brightfield microscopy, go to:
http://imagejdocu.tudor.lu/imagej-documentation-wiki/how-to/how-to-correct-background-
illumination-in-brightfield-microscopy

Filters, enhancement, etc.

Don’t forget to use these tools if necessay, e.g. Process – Enhance contrast or Median or Unsharp
Mask filter, to improve ability to segment the colours.

Acknowledgements
Thanks to the following people for providing images or slides:
Lorraine Rolston
Brent Beaumont
Ursula Byrne
Claire French
Suzanne Reid
Susanne Franke

Thanks also to Dr. Gabriel Landini (School of Dentistry, The University of Birmingham, for
clarification of the Colour Deconvolution plugin.

23