texture analysis tools?

[Kenneth Sloan-2]

I need to do some texture analysis.  Features range in size from 1-5 pixels wide.  Ideally, I would like to:

a) use 25x25 pixel blocks (centered on every pixel in a 1500x1500 image)
b) characterize the texture present in each 25x25 block

I’m probably interested in using large-scale textures to DISQUALIFY a 25x25 block from further analysis, and measure small-scale texture in 25x25 blocks that appear uniform at the macro level.
I’d be perfectly happy to generate many different derived images, representing a wide range of texture “dimension”.

But…I’d rather not re-invent the world.  I haven’t done any serious texture work in over 20 years, so I’m woefully out of touch.  I’m looking for pointers to:

a) a good review article describing the current state of theory, and
b) (if available) an ImageJ set of tools designed specifically for this kind of analysis.

All clues gratefully rented.

[currently, I’m following previous work that used coefficient of variation (standard deviation divided by mean) over these 25x25 blocks.  But, this is confounded by regions where the 25x25 block straddles two very distinct regions in the image.  I’m trying to improve things by rejecting blocks which show large-scale patches.  Ideally, I want to measure high-frequency textures and ignore low-frequency patterns.  At the moment, I’d be satisfied to simply reject blocks that exhibit low-frequency patterns, but with the right set of texture descriptors, I might be able to do slightly better.  I’m not (yet) ready to do Fourier analysis on the blocks - I’d prefer to stay in the spatial domain - when I am ready, I’d have no problem moving to 32x32 blocks]

--
Kenneth Sloan
[hidden email]

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Herbie-4]

Dear Kenneth,

local auto-correlation comes to my mind.

Best

Herbie

:::::::::::::::::::::::::::::::::::::::::::
Am 02.03.15 um 04:37 schrieb Kenneth Sloan:

> I need to do some texture analysis.  Features range in size from 1-5
> pixels wide.  Ideally, I would like to:
>
> a) use 25x25 pixel blocks (centered on every pixel in a 1500x1500
> image) b) characterize the texture present in each 25x25 block
>
> I’m probably interested in using large-scale textures to DISQUALIFY a
> 25x25 block from further analysis, and measure small-scale texture in
> 25x25 blocks that appear uniform at the macro level. I’d be perfectly
> happy to generate many different derived images, representing a wide
> range of texture “dimension”.
>
> But…I’d rather not re-invent the world.  I haven’t done any serious
> texture work in over 20 years, so I’m woefully out of touch.  I’m
> looking for pointers to:
>
> a) a good review article describing the current state of theory, and
> b) (if available) an ImageJ set of tools designed specifically for
> this kind of analysis.
>
> All clues gratefully rented.
>
> [currently, I’m following previous work that used coefficient of
> variation (standard deviation divided by mean) over these 25x25
> blocks.  But, this is confounded by regions where the 25x25 block
> straddles two very distinct regions in the image.  I’m trying to
> improve things by rejecting blocks which show large-scale patches.
> Ideally, I want to measure high-frequency textures and ignore
> low-frequency patterns.  At the moment, I’d be satisfied to simply
> reject blocks that exhibit low-frequency patterns, but with the right
> set of texture descriptors, I might be able to do slightly better.
> I’m not (yet) ready to do Fourier analysis on the blocks - I’d prefer
> to stay in the spatial domain - when I am ready, I’d have no problem
> moving to 32x32 blocks]
>
> -- Kenneth Sloan [hidden email]
>
> -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[audrey karperien-3]

Kenneth, the FracLac plugin does texture analysis over blocks of whatever size you specify for lacunarity, fractal/multifractal dimension, and density for binary or grayscale images.  If it suits your purposes in general but you need the output modified, flip me an email and I'll at least give it a shot to try to get it doing what you need.
Regards,Audrey Karperien
 


      From: Kenneth Sloan <[hidden email]>
 To: [hidden email]
 Sent: Sunday, March 1, 2015 10:37 PM
 Subject: texture analysis tools?
   
I need to do some texture analysis.  Features range in size from 1-5 pixels wide.  Ideally, I would like to:

a) use 25x25 pixel blocks (centered on every pixel in a 1500x1500 image)
b) characterize the texture present in each 25x25 block

I’m probably interested in using large-scale textures to DISQUALIFY a 25x25 block from further analysis, and measure small-scale texture in 25x25 blocks that appear uniform at the macro level.
I’d be perfectly happy to generate many different derived images, representing a wide range of texture “dimension”.

But…I’d rather not re-invent the world.  I haven’t done any serious texture work in over 20 years, so I’m woefully out of touch.  I’m looking for pointers to:

a) a good review article describing the current state of theory, and
b) (if available) an ImageJ set of tools designed specifically for this kind of analysis.

All clues gratefully rented.

[currently, I’m following previous work that used coefficient of variation (standard deviation divided by mean) over these 25x25 blocks.  But, this is confounded by regions where the 25x25 block straddles two very distinct regions in the image.  I’m trying to improve things by rejecting blocks which show large-scale patches.  Ideally, I want to measure high-frequency textures and ignore low-frequency patterns.  At the moment, I’d be satisfied to simply reject blocks that exhibit low-frequency patterns, but with the right set of texture descriptors, I might be able to do slightly better.  I’m not (yet) ready to do Fourier analysis on the blocks - I’d prefer to stay in the spatial domain - when I am ready, I’d have no problem moving to 32x32 blocks]

--
Kenneth Sloan
[hidden email]

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

   

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html