Threshold preview Display Inconsistencies

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Threshold preview Display Inconsistencies

Olivier Burri
Hi all,

We are having a problem with the threshold display for 32 bit images.

Example.
An image with a wide range of values, we set the threshold to 1-Infinity, the whole image appears in red, but when we hit Apply and set "Background to NaN", the proper pixels are set to NaN on the image, making what the preview was showing inconsistent.

We had a worse problem where an image was rescaled to fit in the 0-1 range of values. The Huang method was giving a preview that looked promising but upon hitting Apply, the shape of the mask vas very different from the one that the preview.

What is behaving differently when we hit preview and when we actually perform the create mask operation in 32-bit images?

Best

Oli

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

Re: Threshold preview Display Inconsistencies

Gabriel Landini
On Wednesday 04 Mar 2015 12:40:11 Burri Olivier wrote:
> We are having a problem with the threshold display for 32 bit images.
>
> Example.
> An image with a wide range of values, we set the threshold to 1-Infinity,
> the whole image appears in red, but when we hit Apply and set "Background
> to NaN", the proper pixels are set to NaN on the image, making what the
> preview was showing inconsistent.

I think this might have to do with the way that 32bit images are handled when
applying the threshold methods (which expect a histogram).
The various thresholding methods of the Auto Threshold plugin which were later
ported into the built-in thresholder are for discrete values (i.e. 8 or 16 bit
images).

I seem to remember that Wayne implemented a way round to apply this to 32bit
images in IJ, so they get pre-binned into a range of values (not exactly sure)
I think it is 256 but cannot remember if it is fixed either.

Cheers

Gabriel

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