Multiple object tracking in binary images

Hello everyone,

I'm currently working on an image processing project. Here is the topic : I need to plot the centroid location as a function of time of different objects (human cells) from multipe images.

I binarise each images and identify the different patterns (using the bwlabel function). The centroids are obtained with the regionprops function. However, the objects are moving within the time so the labels are changing as well.

Once the first image has been labeled, I need to keep the same values (of the labels) for all the next images.

Here are two attached files, I hope it will be clearer. On the image "Objects1", the white circle is labeled as '1' but on the second one, it moved so the label changed (it's '2' now). Assuming that "Objects1" is the first image, I just need the circle to have the same label ('1') all the time.

I'm sorry if it's a bit confusing, I've been doing some research but couldn't find any answer.

Thanks!

Kind Regards,

Adrian

Answer 1

[Bruno Pop-Stefanov]

Instead of using bwlabel, you could use bwconncomp to find your components and then label them yourself based on the number of pixels that each component contains. It would work if no two components have the same size. For example, if you identify the disk to have N pixels, then you can search through your list of components in each frame for a component of size N pixels and always label it 1. Good luck!

[Adrian]

It sounds like a good idea, I'll try this out!

Thanks :-)

Answer 2

[Adrian]

Hey,

I'm still struggling with this issue. Bruno's propositation looked good but it does not work actually. Maybe anyone has some other solutions?

Thanks!

[Image Analyst]

Why didn't it work? Are there two objects with the same number of pixels? If so you might have to look at other features such as perimeter length.