Contents || Index
05/02/2008


Image Segmentation using Lispix

Illustration of "Stitching" and "Joining" - Bottom Segment of Blob Tool

Before Lispix can extract any morphological information from the images, the images must be segmented – that is, the computer must be told what is particle, what is annotation, and what is background.  This is usually obvious to the eye, but not so obvious to the computer.  Images are segmented when the pixels are assigned either to background or objects of interest.

Lispix is used to segment the images of particles, with the following steps:

The examples that follow illustrate these steps.



 Segmentation Example 1:  Typical SEM Image of a Particle

Preprocessing

This image is highly shaded. Thresholding fails miserably.


Sliding the left part of the red bar in the threshold slider selects more of the bottom of the particle, but the top is drowned in the background. Flattening takes care of this:


The flattened image looks like this:

Some of the "3D" shape of the particle is now gone, but the background has more uniform intensity - thresholding will reasonably outline the particle:


Threshold the image using either:

Now, blob the image:

The particle is not outlined at all.

Chances are that the blob containing the particle 'leaks' over to the edge of the image at some point.

Also, note here that the blobs that were found are considerably smaller than the particle. Avoid excluding these by hand later by increasing the minimum blob size:

It looks as if this outline follows most of the visual particle outline. To examine and clean it up:

 

Excluding Unwanted Blobs


The yellow line (blob outline) deviates from the particle boundary at two points (red arrows).

To better show the procedure, zoom the image x4. Zooming occurs about the center of the image. So as to not get 'lost',


The circle below marks the gap where the outline diverges to the interior of the particle.


Sometimes the gap is easier to see by plotting the averaged outline:

Stitching a Gap in the outline.


Replot the actual perimeter, and click on either side of the gap to 'stitch' it up:


Clicking 'rev' shows either half of the now split outline.

This is the desired (exterior) half.



Click 'rev' so that the exterior part is shown, and then 'stitch' to close up the gap.

Note: Be sure to use rev! a time or two BEFORE 'stitch' to make sure you are keeping the desired piece of the outline. 'Undo!' might work in case you don't. Also, if an outline requires work, saving versions periodically might be prudent: blob / save to file.


Now the exterior excursion(s) of the outline remain. There is some ambiguity at the upper red arrow.

Showing the smoothed outline resolves the ambiguity:

Clicking on each side of the neck and stitching results in the pretty much cleaned up outline.


Note - dragging or zooming the image will erase the outline. Redraw by clicking on the appropriate button.


How much more of the outline to clean up is a matter of judgment. For example, the straight edges marked in red look to me to be obscured by collodion:

The stitch function draws a straight line between the selected points, which is appropriate in this case. Otherwise, the 'drw!' button allows manual drawing between the selected circles.

Repeated cycling of R! and P! allows examining of other areas of the boundary that might need fixing, such as that within the red oval above. Both of these indentations in the yellow boundary are due to surface cracks or indentations, and not due to a true indentation in the outline of the particle. Fixing these and a few other places results in the cleaned up outline, representative of what is in the outl6 folder.

Smoothing the outline, which is visually more appropriate, can also be done. I did not do this with any of the outlines in the outl6 folder.

Once the outline is smoothed, the new outline can be stored (Blob / Save to file) and used for morphological measurements. This should be done with care, as fine texture or wiggles in the boundary that are real data, might be lost. That is why I did not smooth the boundaries at all. There are instances of particles taken at higher resolutions, where there appears to be a lot of empty magnification. The small wiggles in these boundaries are clearly due to noise, and could be smoothed out.