How to quantify the contrast of particular features in an image?

[Josh Doe-2]

This is somewhat off-topic; if any of you can recommend a good image
analysis mailing list or forum, I would appreciate it, so I can ask these
types of questions there.

I'm writing a plugin for ImageJ that is supposed to give some sort of
contrast number which quantifies how easy or difficult it is to pick out
individual features in an image. The user makes a selection, and I then take
that ROI and the image and produce this number. I have two questions:

1) What metric should I use for this number? What I've found so far are the
difference of the feature and background mean pixel values, and then adding
in the feature variance in quadrature. These don't really satisfy me though.
I feel there should be some higher order statistics involved in this, but I
don't have a good background in image analysis. I've searched Google scholar
and read some papers, but haven't really found anything like I need.

2) What do I consider the background? So far I've been using the Euclidiean
distance map (EDM) plugin, and then using the Wand to choose my background
so it has roughly the same number of pixels as the feature has. This creates
an annulus or band around the feature, much like the "Make Band" command
does. Or maybe I should use the entire image for my background.

If anyone has any pointers, or can tell me where to look, I'd really
appreciate it!

Josh D

[Cabrera, Julio (NIH/NCI) [E]]

Hi Josh,

Have you considered Haralick's contrast parameter?
Take a look at my texture analysis plugin and let me know if it is of any use.

Cheers,

Julio Cabrera


 

-----Original Message-----
From: Josh D [mailto:[hidden email]]
Sent: Wednesday, December 27, 2006 4:02 PM
To: List IMAGEJ
Subject: How to quantify the contrast of particular features in an image?

This is somewhat off-topic; if any of you can recommend a good image analysis mailing list or forum, I would appreciate it, so I can ask these types of questions there.

I'm writing a plugin for ImageJ that is supposed to give some sort of contrast number which quantifies how easy or difficult it is to pick out individual features in an image. The user makes a selection, and I then take that ROI and the image and produce this number. I have two questions:

1) What metric should I use for this number? What I've found so far are the difference of the feature and background mean pixel values, and then adding in the feature variance in quadrature. These don't really satisfy me though.
I feel there should be some higher order statistics involved in this, but I don't have a good background in image analysis. I've searched Google scholar and read some papers, but haven't really found anything like I need.

2) What do I consider the background? So far I've been using the Euclidiean distance map (EDM) plugin, and then using the Wand to choose my background so it has roughly the same number of pixels as the feature has. This creates an annulus or band around the feature, much like the "Make Band" command does. Or maybe I should use the entire image for my background.

If anyone has any pointers, or can tell me where to look, I'd really appreciate it!

Josh D

[Josh Doe-2]

Julio,
That seems to be heading towards the direction of what I'm looking for, as
I've come across GLCM's before. The Haralick contrast seems to be just for
how much contrast an image has. What I need is the contrast of a particular
feature with respect to the background surrounding it. For example, a black
box on a white background would have a contrast of unity, 1.0. As would a
white box on a black background (I only care about absolute values). A gray
box on a noise background (same means)
would have a fairly high contrast. Like I said, I can't find anything
in journals, so I assume I'm just not using the right terminology to
find what I'm looking for.

Thanks though, it is closer to what I'm looking for in regards to including
more than just 1st order statistics.

Josh D

On 12/28/06, Cabrera, Julio (NIH/NCI) [E] <[hidden email]> wrote:

>
> Hi Josh,
>
> Have you considered Haralick's contrast parameter?
> Take a look at my texture analysis plugin and let me know if it is of any
> use.
>
> Cheers,
>
> Julio Cabrera
>
>
>
>
> -----Original Message-----
> From: Josh D [mailto:[hidden email]]
> Sent: Wednesday, December 27, 2006 4:02 PM
> To: List IMAGEJ
> Subject: How to quantify the contrast of particular features in an image?
>
> This is somewhat off-topic; if any of you can recommend a good image
> analysis mailing list or forum, I would appreciate it, so I can ask these
> types of questions there.
>
> I'm writing a plugin for ImageJ that is supposed to give some sort of
> contrast number which quantifies how easy or difficult it is to pick out
> individual features in an image. The user makes a selection, and I then take
> that ROI and the image and produce this number. I have two questions:
>
> 1) What metric should I use for this number? What I've found so far are
> the difference of the feature and background mean pixel values, and then
> adding in the feature variance in quadrature. These don't really satisfy me
> though.
> I feel there should be some higher order statistics involved in this, but
> I don't have a good background in image analysis. I've searched Google
> scholar and read some papers, but haven't really found anything like I need.
>
> 2) What do I consider the background? So far I've been using the
> Euclidiean distance map (EDM) plugin, and then using the Wand to choose my
> background so it has roughly the same number of pixels as the feature has.
> This creates an annulus or band around the feature, much like the "Make
> Band" command does. Or maybe I should use the entire image for my
> background.
>
> If anyone has any pointers, or can tell me where to look, I'd really
> appreciate it!
>
> Josh D
>