Lecture: Image Segmentation Part 1
Slides
Image Thresholding
Open the class introduction slides in a separate window: https://stats4neuro.netlify.com/slides/04-image_segmentation#1
Remember, you can hit p to open up the preview mode and see the slides.
Post-Class
Please fill out the following survey and we will discuss the results during the next lecture. All responses will be anonymous.
- Clearest Point: What was the most clear part of the lecture?
- Muddiest Point: What was the most unclear part of the lecture to you?
- Anything Else: Is there something you’d like me to know?
Muddiest Points
I was a little bit confused about the watershed algorithm. When you are setting the boundaries around objects, are you able to set up variable boundaries? (ie. set up a border between object x and y, which is different from the border between x and z)? - to achieve better separation for example.
The boundaries are determined by when each valley is flooded by another valley - you basically define the boundary right before the flooding (building a dam).
Do these segementations only work you start from a binarized image? and when is it better to use distance transformation versus the watershed algorithm?
Ok. watershed only works on distance mapping, as it has to traverse a continuous surface, not a discontinuous surface like the binarized image.
the 3D representation of the nearest neighbor map. I liked the explanation of watershed algorithms, but I was initially confused what the 3D representation was – this got cleared up in the end though so I think it makes sense now
As far as the watershed segmenting, I’m still a little muddy on the 3D representation
I’m still working on understanding the watershed algorithm and how it is extrapolating a 2D image into 3D.
the plot of the distance matrix didn’t make as much sense - wouldn’t there be a gradient on the exterior of the image also?
Yes. I understand this was confusing. Hopefully the lab will clarify this. The distmap() function is what is creating the 3D surface.
Can you do the watershed segmentation to separate the boundaries and then apply a filter to round out the boundaries and get a better estimate of the area/size?
Because the output of the watershed is a labeled image, it doesn’t make sense to filter after the segmentation, as you will lose the labels for each of the objects.