Color thresholding

[Craig Oppel]

Hello All,

I am working on a project that involves microwave extraction of leaves onto
TLC plates to quantify alkaloids in cardinal flower. The final plates have
a yellow background with a red/orange alkaloid positive leaf imprint. I
have tried color thresholding to eliminate the yellow background, and I can
get acceptable results, but I am concerned because I have to make manual
adjustments to the intensity sliders. I have found that if I first convert
the TIFF images of the TLC plates into 8-bit gray scale images and then use
the adjust contrast/brightness auto function I get really good results.
What I am concerned about is I have no idea how the algorithim processes
the picture. I am taking great pains to eliminate bias in my design. Is
there a way to eliminate the yellow background that would ensure that all
images are treated identically?

Thank you
Craig

[Weller Andrew Francis]

Craig,

One thought with this. Perhaps you can decompose your colour image into
3 separate RGB channels. This will effectively give you 3 grey-scale
images, if you've used 16M colours?! See the ExtractRGBChannels.txt
macro. These 3 channels can then be treated separately when you
(auto?)threshold. They can then be recomposed into one thresholded
colour image with a better result.

Perhaps...? I am relatively new to ImageJ, so I hope it helps!?

Andy

On Fri, 2006-06-23 at 10:17 -0400, Craig Oppel wrote:

> Hello All,
>
> I am working on a project that involves microwave extraction of leaves onto
> TLC plates to quantify alkaloids in cardinal flower. The final plates have
> a yellow background with a red/orange alkaloid positive leaf imprint. I
> have tried color thresholding to eliminate the yellow background, and I can
> get acceptable results, but I am concerned because I have to make manual
> adjustments to the intensity sliders. I have found that if I first convert
> the TIFF images of the TLC plates into 8-bit gray scale images and then use
> the adjust contrast/brightness auto function I get really good results.
> What I am concerned about is I have no idea how the algorithim processes
> the picture. I am taking great pains to eliminate bias in my design. Is
> there a way to eliminate the yellow background that would ensure that all
> images are treated identically?
>
> Thank you
> Craig

[Martin du Saire]

You could also use the 3D Color Inspector to "inspect" the color
distribution in your images.  Then you can decide which color space
will provide the best options for segmentation, processing, etc.

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

M

At 09:17 AM 6/23/2006, you wrote:
>Hello All,
>
>I am working on a project that involves microwave extraction of leaves onto
>TLC plates to quantify alkaloids in cardinal flower. The final plates have

[Dimiter Prodanov]

Hi Craig,

I have a different suggestion. First, single global threshold is good ONLY for
1 channel (ie. gray-scale) images. To work around
this you can convert tour color space
to CMYK - there is a separate yellow channel.
Make the statistics of the histogram
and then subtract the desired value of the
Y-channel. Then you can convert back to
RGB and see /measure the outcome. Another
possibility is to use the Color Deconvolution
of Gabriel Landini, but it will not save you from
the statistics because you have to
adjust the settings of the plugin to separate well the yellow background.
About the subjectivity: making the image statistics and levering the images to
have similar values; i.e. equal mean intensity of
some of the channels will give you
some way out.

best regards


Dimiter



_______________________________________________________________________
Dr Dimiter Prodanov, MD, Ph.D.

Neural Engineering Rehabilitation Laboratory
(Laboratoire de Génie de la Réhabilitation Neurale)
Département de Physiologie et Pharmacologie
Université catholique de Louvain
Avenue Hippocrate, 54
POBox UCL-5446 / B-1200 Bruxelles -Belgique-
Phone: [phone]
Fax: [phone]

http://www.md.ucl.ac.be/gren