Segmenting bubbles in a microchannels

Hi

I'm trying to process some images of bubbles in microchannels. I will be recording high speed videos and processing each frame to get an understanding of the behavior of the bubbles. I will be looking at the bubble movement velocity (and direction), growth rate, radii of curvature (which I'm not sure I know how to do), and contact angles. At the moment however I' having trouble processing the images. The images are not very good, but I think I should be able to process them in a better way.

I'm using one background image which is an image of the channel without any flow and another image which has flow. The idea is to subtract the blank image from the flow image so that in a perfect world all that would remain would be the bubbles! However, lighting and other issues have made this difficult.

The code I'm using is like this:

Im1 = B00012; %image of the device without flow
Im1A = imclose(Im1, strel('diamond',27));
%the only reason I'm using diamonds as the structural element is that it
%seemed to work ok. The purpose here is to remove the device from the image
%to create a background image.
Im1B = Im1A - Im1; %subtracting the background image.
Im1 = imadjust(Im1B);

Im2 = B00633; %This image shows the device with bubbles flowing through.
Im2A = imclose(Im2, strel('diamond', 27));
Im2B = Im2A - Im2;
Im2 = imadjust(Im2B);

Im3 = Im2 - Im1;
%subtractin the bubble-less image in hope of getting an image containing the bubbles only.

imshow(Im3)

I = imopen(Im3, strel('disk', 2)); %an attempt to remove the extra pixels.

I2 = im2bw(I,0.27);

figure, imshow(I2)

I3 = bwconvhull(I2, 'objects', 8);
figure, imshow(I3)
%trying to fill in the bubbles. I was kind of looking for a way to
%reconstruct the bubbles which were damaged during thresholding. This image
%is not accurate enough for my purposes. Especially since bubbles in
%separate channels have merged. I think it would be best to only apply conv
%hull to the header area (not the channels)...

Links to the original images:

The device without flow https://dl.dropboxusercontent.com/u/45041086/B00012.tif

The device with flow (the blobs in the channels are the bubbles) https://dl.dropboxusercontent.com/u/45041086/B00633.tif

The images are about 10 MB each. Sorry about the large size. If you run my code loading these images into MATLAB with their current names you should get the same results I do.

Any suggestions would be highly appreciated. Thank you.

Answer 1

[Image Analyst]

Why can't you just increase the camera exposure time? Can you also post the intermediate and final images? Because I can't download them unless I buy Quicktime Pro (according to when I right click on the image).

[Micofludics]

Hmmm, the pictures just start downloading when I click on them. Their TIF images so you should be able to open them with pretty much anything.

The exposure time is something I'm going to look into. I don't capture the images, I only process them. I've been told they can't get sharp images at higher exposure times but I'm not sure how true that is. I just use imadjust to get a better image, but I never really thought about whether using the full bandwidth would help my segmentation. Could that have a significant impact?

Here is the first intermediate step (Im3): https://dl.dropboxusercontent.com/u/45041086/Im3.tif

The next one is from after the thresholding (I2): https://dl.dropboxusercontent.com/u/45041086/I2.tif

The last one is what I get after I use bwconvhull (I3): https://dl.dropboxusercontent.com/u/45041086/I3.tif

Thanks

[Image Analyst]

I still can't see the first one. Can you try snag.gy?

[Micofludics]

I hope this helps:

Im3: http://snag.gy/SIltt.jpg

I2: http://snag.gy/9GHne.jpg

I3: http://snag.gy/Z5NZu.jpg

[Image Analyst]

Yes, much better. Since it looks like your slide is in some kind of jig to accurately reposition it, why can't you just use fixed coordinates to examine the gray levels to determine how long the path is?

[Micofludics]

I don;t understand what you mean. Can you explain a bit more? By path do you mean length of bubble? How do I examine the gray levels? I was under the impression that the only way I can measure the features I need would be to separate the bubbles out to study them individually. Although I'm not sure how to find the radius of curvature of the bubbles either...

I also noticed that if I use a disk structural element for the 'close' morphological operation the bubbles in the channels disappear but the ones in the header area (the ones which are on the left side) actually become better. Do you have any idea on how I can apply different operations to different parts of the image?