Bark texture analysis

[Chantel Davies]

Dear Listers,

I need advice on how to conduct a texture analysis of tree bark (trembling
aspen, or Populus tremula).

I have a set of 300 raw images taken at a specific distance from each tree,
with a region of interest of 500x500 pixels selected for the analysis.
The trees are all the same species but grow in clones which have differences
in certain physical characteristics - one of which seems to be bark texture.
 I can recognise these differences by eye, but I would like to quantify the
differences in a sound and scientifically robust way.

Since I am new to imageJ and digital image analysis, I don't really know how
I should proceed, and I have read of a range of methods for this type of work.
Would anyone have suggestions or recommendations on how to begin?  I believe
I need to standardise the images first, then do an analysis of some
description - any ideas?

Best wishes,

Chantel Davies

[karo03]

Hi, texture analysis is always difficult since it is mostly not clear  
what to measure, not only for novices. One reason might be the lacking  
description of the differences to detect.
However, one way to tackle the problem is to try to "peel off" the  
texture by certain filter operations. Try first with smoothing and  
looking at the difference (the residue), then with non-linear  
operations e.g. non-linear Gaussian or opening/closing (on grey  
images, not binary), always considering the residue. Opening/closing  
is helpful to "peel off" either the dark or the bright parts, which  
might be helpful for bark.
Texture analysis is still a problem!
Best regards
Karsten

Am 06.02.2009 um 15:23 schrieb Chantel Davies:

> Dear Listers,
>
> I need advice on how to conduct a texture analysis of tree bark  
> (trembling
> aspen, or Populus tremula).
>
> I have a set of 300 raw images taken at a specific distance from  
> each tree,
> with a region of interest of 500x500 pixels selected for the analysis.
> The trees are all the same species but grow in clones which have  
> differences
> in certain physical characteristics - one of which seems to be bark  
> texture.
> I can recognise these differences by eye, but I would like to  
> quantify the
> differences in a sound and scientifically robust way.
>
> Since I am new to imageJ and digital image analysis, I don't really  
> know how
> I should proceed, and I have read of a range of methods for this  
> type of work.
> Would anyone have suggestions or recommendations on how to begin?  I  
> believe
> I need to standardise the images first, then do an analysis of some
> description - any ideas?
>
> Best wishes,
>
> Chantel Davies

Karsten
[hidden email]

[Joris FA Meys]

Hi,

it's an interesting problem, and I believe it is a bit dependent on the
nature of the differences in texture. For Populus tremula, I believe it is
basically about the size, number and/or position of the lenticels on the
bark. Filters will have difficulties with finding the lenticels due to the
patchiness of the bark underneath. One way is to point them out by hand,
e.g. by using a circular ROI and adding them all to a roi manager.

You can use the Multi Measure plugin for that.(included in the MBF version
of ImageJ : http://www.macbiophotonics.ca/imagej/ or at
http://www.optinav.com/Multi-Measure.htm). Select a lenticel, click Add&draw
so you see which ones you have selected already. Make a new blank image (eg.
by duplicating and setting the brightness to maximum). Then you can use the
fill option to draw your selections on this new image. Treshold to make it
binary and then you can use any kind of particle analysis for getting the
measures you want.

It's maybe not the most elegant method, but it should get you somewhere.

Kind regards
Joris

On Sat, Feb 7, 2009 at 10:35 AM, Karsten Rodenacker <[hidden email]>wrote:

> Hi, texture analysis is always difficult since it is mostly not clear what
> to measure, not only for novices. One reason might be the lacking
> description of the differences to detect.
> However, one way to tackle the problem is to try to "peel off" the texture
> by certain filter operations. Try first with smoothing and looking at the
> difference (the residue), then with non-linear operations e.g. non-linear
> Gaussian or opening/closing (on grey images, not binary), always considering
> the residue. Opening/closing is helpful to "peel off" either the dark or the
> bright parts, which might be helpful for bark.
> Texture analysis is still a problem!
> Best regards
> Karsten
>
> Am 06.02.2009 um 15:23 schrieb Chantel Davies:
>
>  Dear Listers,
>>
>> I need advice on how to conduct a texture analysis of tree bark (trembling
>> aspen, or Populus tremula).
>>
>> I have a set of 300 raw images taken at a specific distance from each
>> tree,
>> with a region of interest of 500x500 pixels selected for the analysis.
>> The trees are all the same species but grow in clones which have
>> differences
>> in certain physical characteristics - one of which seems to be bark
>> texture.
>> I can recognise these differences by eye, but I would like to quantify the
>> differences in a sound and scientifically robust way.
>>
>> Since I am new to imageJ and digital image analysis, I don't really know
>> how
>> I should proceed, and I have read of a range of methods for this type of
>> work.
>> Would anyone have suggestions or recommendations on how to begin?  I
>> believe
>> I need to standardise the images first, then do an analysis of some
>> description - any ideas?
>>
>> Best wishes,
>>
>> Chantel Davies
>>
>
> Karsten
> [hidden email]
>