As usual, my approach would involve writing Java code. Computing a “length” of a perimeter is tricky, because it depends on the size of the pixels. Consider the famous paper “How long is the coastline of Great Britain?”
My first attempt would be to count the BLACK pixels that have at least 1 WHITE neighbor.
This isn’t quite right, but I would validate this against known images to see if it matters, to you. The next twiddle might be to count a BLACK pixel with 1 WHITE neighbor as contributing 1.0 to the length, and BLACK pixels with more than 1 WHITE neighbors as contributions 1.4 to the length. I’m not even sure that that’s correct - but, again, I would try it on known shapes to see if it’s “close enough for gov’t work”.
There may be a clever way to do this as a macro, but it’s so simple to do in Java that I wouldn’t look any further.
There are more complicated ways to do this - that probably give more precise answers. If necessary, I would crawl around the perimeter, selecting boundary pixels and then imposing a distance constraint (don’t try to count every pixel, because that gives unwanted tiny-scale pieces - perhaps require the next point to be something like 5 pixels distant from the previous one. That is - automatically sketch the boundary, and compute the length of the sketch. The image you showed looks amenable to this approach.
The downside is: this is a moderately difficult Java program to write, if you haven’t written a lot of similar ones, before. I would classify it as a reasonable class project for an undergraduate CS major (writing it, debugging it, validating it, wrapping it up with proper documentation, etc.).
> On Feb 26, 2016, at 03:22 , Herbie <
[hidden email]> wrote:
>
> Dear Dave,
>
> I must admit that Michael's length estimate is much better than the simple one that I've proposed. Mine may be regarded as a a lower bound estimate...
>
> Best
>
> Herbie
>
> ::::::::::::::::::::::::::::::::::::::::::::
> Am 25.02.16 um 21:36 schrieb Michael Schmid:
>> Hi David,
>>
>> you can use the wand tool to get the outline of the airspace, then
>> measure the perimeter and subtract the length at the image border.
>> Unfortunately this will be very inaccurate because of the pixelated
>> edge; the wand will follow the pixel outlines.
>>
>> Better have a short macro that takes the coordinates with
>> getSelectionCoordinates, then eliminate those at the border
>> (typically, the last 3 points, unless you have a particle touching
>> the 'air' borders). Then create a new segmented line with these
>> coordinates or a subset of them (e.g. each 2nd or 3rd one, or moving
>> averages, for some smoothing); see the makeSelection command with
>> type "polyline".
>>
>> A very simple version of the macro, without smoothing, is like this,
>> but it will follow the pixel outlines:
>>
>> setAutoThreshold("Default"); run("Convert to Mask"); //make sure it's
>> a binary image doWand(0, 0, 0, "Legacy"); //assumes (0,0) is air
>> getSelectionCoordinates(x,y); newLength = x.length-3; x2 =
>> Array.trim(x, newLength); y2 = Array.trim(y, newLength);
>> makeSelection("Polyline", x2, y2); run("Set Measurements...",
>> "perimeter redirect=None decimal=3"); run("Measure");
>>
>>
>> Michael
>> ________________________________________________________________ On
>> Feb 25, 2016, at 18:06, David A. Schneider wrote:
>>
>>> Greetings,
>>>
>>> I need to process a large image set of masks (like the one
>>> attached) which represent a tissue cross section (lower white
>>> region) and surrounding airspace (upper black region). I've written
>>> macros to get me this far but I don't know if there is already an
>>> existing automatic way to measure the distance of the surface.
>>>
>>> Note, as in this example image, that there are occasional
>>> white-black interfaces not to be included. These occur in the
>>> non-tissue space (in the image the single white blotch in upper
>>> right quadrant) and those holes in the deeper tissue space (one
>>> about mid-lower left side, another on low right side). All images
>>> are oriented in this way to start. I'm not concerned about not
>>> catching the bit of surface that is just below the left edge
>>> boundary. If it will make for a much simpler solution, I could
>>> remove the bits of white-in-black and black-in-white not
>>> representing the surface to be measured. But as you can see, hand
>>> drawing the surface with a line for a hundred or more images is not
>>> feasible nor likely to be accurate.
>>>
>>> Is there a plugin or function already out there that I can use
>>> (preferably in a macro or batch process) to get this done?
>>>
>>> Thank you for your assistance,
>>>
>>> Dave
>>
>> -- ImageJ mailing list:
http://imagej.nih.gov/ij/list.html>>
>
> --
> ImageJ mailing list:
http://imagej.nih.gov/ij/list.html
Locked
Measure length of surface white boundary.tif (872K)