I am dealing with images from biological slides and want to segment out the ROI (cells) from the background. I have tried k-means clustering, otsu thresholding, adaptive thresholding, edge detection etc. Running out of ideas now.
Which other segmenting techniques can be tried for such images? Keeping in mind that the color of cells and background can change
click here for image
$\begingroup$
$\endgroup$
Add a comment
|
3 Answers
$\begingroup$
$\endgroup$
Some very cool although a bit more time consuming options:
Active contours will try to find the best "edge" for the cells while maintaining smoothness. and is implemented in matlab2013a ( active contour)
You might try using mser which is an approach which detects regions which are similiar and in color and stable over the image. which means they stay about the same size over a wide array of thresholds. I used it for a similar problem with detecting ultrasound tumors.
If you have a good database of images ( good = covering most cases ) I would try some machine learning algorithms. for example, you could learn on the mean and variance of the objects in the image.
If you have trouble separating the cells after segmentation ( i.e. you have two connected cells ) I would try looking at the watershed function.
$\begingroup$
$\endgroup$
Microsoft has developed Branch-and-MinCut, and well applied it to medical images. The paper is here: http://research.microsoft.com/en-us/um/people/ablake/papers/ablake/image%20segmentation%20eccv2008.pdf
It is a quite good algorithm, as it address global minima.
Furthermore, they released it open source: http://research.microsoft.com/en-us/downloads/994e80f7-007b-4fed-a6e9-db5f4f32ccd1/
answered Jan 8, 2014 at 8:42
$\begingroup$
$\endgroup$
You can try with the mean shift algorithm. It is implemented in Orfeo Toolbox.
