isolating lichens from the background

[Charles R Parker]

I'm new to image processing and ImageJ, so please bear with me. I need to
be able to measure the area of healthy and necrotic lichens in
photographs. Examples are at http://dlia.org/images/Parker/ParkerTIFS.html

 So far, I have not been able to isolate the lichen from the background,
let alone measure anything. I would appreciate any suggestions anyone may
have on a suitable approach.

Thanks,
Chuck
=========================
Charles R. Parker, Ph.D.
Research Aquatic Biologist
U. S. Geological Survey
1316 Cherokee Orchard Road
Gatlinburg, TN 37738

E-mail: [hidden email]
Phone: [phone]

[Albert Cardona]

 > I'm new to image processing and ImageJ, so please bear with me. I need to
 > be able to measure the area of healthy and necrotic lichens in
 > photographs. Examples are at
http://dlia.org/images/Parker/ParkerTIFS.html
 >
 >  So far, I have not been able to isolate the lichen from the background,
 > let alone measure anything. I would appreciate any suggestions anyone may
 > have on a suitable approach.


An idea: perform texture analysis. The lichens look fairly homogeneous
texture-wise.

A very crude texture analysis: for a given circular ROI, determine by
whatever statistics (color, particle count, distance between local
maxima, ..., or a combination of them) whether it belongs to the lichen
or not. Then iterate the ROI over entire image, centered on each pixel,
so that you an create a mask: white for lichen pixels, black for
non-lichen pixels.

I am sure the literature has better ways to do it. This looks like the
kind of problem a neural network should be able to get trained for.

Hope that helps.

Albert

--
Albert Cardona
http://www.mcdb.ucla.edu/Research/Hartenstein/acardona

[Charles R Parker]

Thank you for the suggestions. I will look into texture analysis and
neural networks.

Chuck
=========================
Charles R. Parker, Ph.D.
Research Aquatic Biologist
U. S. Geological Survey
1316 Cherokee Orchard Road
Gatlinburg, TN 37738

E-mail: [hidden email]
Phone: [phone]



Albert Cardona <[hidden email]>
Sent by: ImageJ Interest Group <[hidden email]>
05/16/2008 10:25 AM
Please respond to
ImageJ Interest Group <[hidden email]>


To
[hidden email]
cc

Subject
Re: isolating lichens from the background






 > I'm new to image processing and ImageJ, so please bear with me. I need
to
 > be able to measure the area of healthy and necrotic lichens in
 > photographs. Examples are at
http://dlia.org/images/Parker/ParkerTIFS.html
 >
 >  So far, I have not been able to isolate the lichen from the
background,
 > let alone measure anything. I would appreciate any suggestions anyone
may
 > have on a suitable approach.


An idea: perform texture analysis. The lichens look fairly homogeneous
texture-wise.

A very crude texture analysis: for a given circular ROI, determine by
whatever statistics (color, particle count, distance between local
maxima, ..., or a combination of them) whether it belongs to the lichen
or not. Then iterate the ROI over entire image, centered on each pixel,
so that you an create a mask: white for lichen pixels, black for
non-lichen pixels.

I am sure the literature has better ways to do it. This looks like the
kind of problem a neural network should be able to get trained for.

Hope that helps.

Albert

--
Albert Cardona
http://www.mcdb.ucla.edu/Research/Hartenstein/acardona

[Wilfred L. Guerin]

I will see if any of our extrapolation and signals analysis systems
are effective on your data, we had quite a round pulling 3d structure
data out of common images of particulate materials like granite or
coral sculpture (where pixel color values at depth are identical
mathematically to particles locally) and in crystals. Also, we have
had good (huge) results with orgaic system correlation, namely
identification of trees or plants in environmental modeling by reverse
modeling known (hint plant identification lookup books)
characteristics and correlating statistically their structure. (system
handles abstract data set creation with physically proven generators
(plant growth simulation) to accurately model current and
historical/future organic composition.)

In your case, most flat plants have a fairly standard growth pattern,
obviously using this as the differential bounds to identify the finite
size and density of growth is suitable, otherwise, isolate solids
statistically and search color based extensions or groupings.

Either way, fuzzy things represented as pixles are never very accurate.

If you control the image collection, it is very easy to use a few
plastic spheres (balls) and a few images of each target area to
exponentiate a 3d representation of your target, thus allowing for
accurate isolation and modeling of growth.

For ij in 2d, statistical color histograms (ired/uv camera) to isolate
targets, filters, then accumulate particles with certain localization
affinity.

Patterns in the sand ;)

Counting particles was never hard!


On 5/16/08, Charles R Parker <[hidden email]> wrote:

> I'm new to image processing and ImageJ, so please bear with me. I need to
> be able to measure the area of healthy and necrotic lichens in
> photographs. Examples are at http://dlia.org/images/Parker/ParkerTIFS.html
>
>  So far, I have not been able to isolate the lichen from the background,
> let alone measure anything. I would appreciate any suggestions anyone may
> have on a suitable approach.
>
> Thanks,
> Chuck
> =========================
> Charles R. Parker, Ph.D.
> Research Aquatic Biologist
> U. S. Geological Survey
> 1316 Cherokee Orchard Road
> Gatlinburg, TN 37738
>
> E-mail: [hidden email]
> Phone: [phone]
>