Advice segmenting a short video

[Pedro J CamelloDr Pedro J Camello]

Hi all,

I need to segment/threshold  a grey video containing a small segment of intestine which develops continuous contractions. The are to be segmented (the intestine) is dark, in a clear background.

Here I have two examples:
https://drive.google.com/file/d/0By5p1UZJKft_c0tCUnNMRkYtMk0/edit?usp=sharing
https://drive.google.com/file/d/0By5p1UZJKft_M0VTV0V1YUd0emM/edit?usp=sharing

My problem is that, when the intestine dilates, its pixel values approaches the value of the backgorund pixels. Therefore this forces me to make a manual thresholding inspecting all the stack to be sure that the threshold limit marks the intestine in all the frames. Though I can do this successfully, it is time consuming (I have to do in hundreds of videos) and what´s worse it introduces subjetive decissions I want to avoid.


Has anybody experience using an automated threshold method for videos or for series of images? The trainable Weka segmentation is too slow, and since the illumination conditions changed from video to video, I should change the training every few videos.

I was thinking of some automated method of segmentation not based on threshold, such as snakes, but I have no experience with it.



Thanks in advance

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

[Anderson, Charles (DNR)]

Pedro,

There is something very strange about your images (ignoring that the checkerboard look of the image does not correspond to pixels in the image). One can move the crosshair slowly over the image and see the grey level change irregularly in ways that do not match the visual image.  Auto threshold and find edges do not find the patterns that look obvious in the images.

But if I change Image | type | 16-bit, then everything seems to behave sensibly.  It appears to work sensibly if changed back to 8-bit as well.

Try converting to 16-bit.   Maybe someone else can provide an explanation.  [My FIJI is not updating without an error message today, so I hope this is not just a failure of my system.]

Charles

-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Pedro J Camello
Sent: Wednesday, February 26, 2014 6:38 AM
To: [hidden email]
Subject: Advice segmenting a short video

Hi all,

I need to segment/threshold  a grey video containing a small segment of intestine which develops continuous contractions. The are to be segmented (the intestine) is dark, in a clear background.

Here I have two examples:
https://drive.google.com/file/d/0By5p1UZJKft_c0tCUnNMRkYtMk0/edit?usp=sharing
https://drive.google.com/file/d/0By5p1UZJKft_M0VTV0V1YUd0emM/edit?usp=sharing

My problem is that, when the intestine dilates, its pixel values approaches the value of the backgorund pixels. Therefore this forces me to make a manual thresholding inspecting all the stack to be sure that the threshold limit marks the intestine in all the frames. Though I can do this successfully, it is time consuming (I have to do in hundreds of videos) and what´s worse it introduces subjetive decissions I want to avoid.


Has anybody experience using an automated threshold method for videos or for series of images? The trainable Weka segmentation is too slow, and since the illumination conditions changed from video to video, I should change the training every few videos.

I was thinking of some automated method of segmentation not based on threshold, such as snakes, but I have no experience with it.



Thanks in advance

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


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

[Pedro J CamelloDr Pedro J Camello]

Hi Charles,

I´ve been unable to reproduce the behaviour you describe. The images are very small (abount 100 x 30 pixels) so that pixels are perfectly differentiated. I have tried to change from 8 to 16 and 32 bits but I see no difference. I have even made a Plot Profile for a vertical line in my original and in 16 and back-to 8 bit images and the plots are the same.
After the conversion I still have the same problem with automated segmentation/thresholding (in fact, in my first attempts I used 32 bit images)

Thanks for your reply

Pedro

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

[Anderson, Charles (DNR)]

My error. I let a default program open your images and then saved them; this changed their size and created the weird behavior I encountered. After saving your images correctly, I tried a few things.  

Adjust | AutoThreshold (turning off the white objects on dark background) suggests some threshold methods that might work better than others (Yan looks Ok on each image).  Trying a Gaussian blur with sigma =1 or 2 before using AutoThreshold may help.

You might apply a stronger blur, then find the minima representing the center of the constricted regions, then track movement of that point to estimate velocity and frequency of peristaltic movements, as a different approach.

There must be methods for background segmentation for videos, but I have no experience with them. Good luck.

Charles

________________________________________
From: ImageJ Interest Group [[hidden email]] on behalf of Pedro J Camello [[hidden email]]
Sent: Saturday, March 01, 2014 9:04 AM
To: [hidden email]
Subject: Re: Advice segmenting a short video

Hi Charles,

I´ve been unable to reproduce the behaviour you describe. The images are very small (abount 100 x 30 pixels) so that pixels are perfectly differentiated. I have tried to change from 8 to 16 and 32 bits but I see no difference. I have even made a Plot Profile for a vertical line in my original and in 16 and back-to 8 bit images and the plots are the same.
After the conversion I still have the same problem with automated segmentation/thresholding (in fact, in my first attempts I used 32 bit images)

Thanks for your reply

Pedro

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