Message Boards Message Boards

1
|
9008 Views
|
5 Replies
|
5 Total Likes
View groups...
Share
Share this post:

How Does ImageFeatureTrack Work? Can it Track Shapes?

Posted 8 years ago

I have been trying out ImageFeatureTrack on some images of cells growing over time (images acquired via timelapse microscopy) and it looks promising. However, sometimes points don't get tracked as expected.

i.) Can anyone explain in simple terms how the features (really points) are tracked and how to improve choices of which points are likely to be successfully tracked?

ii.) Is there any way to combine the points into contours (i.e. a shape rather than just a collection of individual points) to be tracked over time?

iii.) Are there other functions in Mathematica that can find a user-input shape (in particular the boundary of the shape) of interest in an image? Sort of like face detect but more generic?

Thanks,

Kosmut

POSTED BY: Kosmut Chodik
5 Replies
Posted 8 years ago

Maybe try MorphologicalPerimeter before ImageFeatureTrack?

http://reference.wolfram.com/language/ref/MorphologicalPerimeter.html

POSTED BY: Dan Von Kohorn

Hi,

(i) As of version 10.4.1, ImageFeatureTrack uses "corners" to track positions across several images. This is similar to https://users.cs.duke.edu/~tomasi/papers/shi/TR_93-1399_Cornell.pdf Conceptually, the corners are detected as in ImageCorners, and tracked on successive images by best fitting the match of local regions around each corner.

(ii) Not at the moment

(iii) That's a great suggestion --- future versions of Mathematica will have tools to make such applications a bit easier. It would be nice if you could find an implementation of Active Shape Models or Snakes in Mathematica in the meanwhile. As a side note, the function ChanVeseBinarize is pretty good at fitting a contour to objects for binary segmentation.

POSTED BY: Matthias Odisio
Posted 8 years ago

Hi Matthias,

Can you provide any links to implementation of Active Contours in Mathematica? I searched on Wolfram Demonstrations and didn't see anything. Also, can you provide any further documentation or examples or how to use ChanVeseBinarize beyond what's available in the Mathematica help? I have tried to use ChanVeseBinarize with a white mask that has been generated from broadly cropping around an object of interest and I couldn't seem to get any better results than LocalAdaptiveBinarize. My expectation was that ChanVeseBinarize would shrink the cropped mask to closely fit the contour of the object of interest but I don't see this happening. Any tips?

Thanks,

Kosmut

POSTED BY: Kosmut Chodik
POSTED BY: Matthias Odisio
Posted 8 years ago

My wild guess would be it's based on SURF or SIFT methods, like ImageKeypoints.

POSTED BY: Jari Kirma
Reply to this discussion
Community posts can be styled and formatted using the Markdown syntax.
Reply Preview
Attachments
Remove
or Discard

Group Abstract Group Abstract