2
$\begingroup$

I have an ultrasonic image represented by a point cloud as shown below:

enter image description here

I need to extract features from this image. These features can be easily recognised by a human brain since they form continuous lines/curves. I am currently using a modified density based clustering algorithm to extract the features. A result is shown below:

enter image description here

where points inside those black borders are extracted.

However, the result is not perfect. If I zoom in at one of the clusters, you can see the points actually form two curves/lines in the cluster, one inside the red rectangle and the other inside the blue rectangle.

enter image description here

For this image, I can apply some different parameters for db-clustering algorithm locally to sub-divide the cluster but if I were to process hundreds of images, this became impossible.

Hence, I am asking if there is any algorithm that can recognise the lines/curves in a point cloud like such.


Update:

Link to sample data

$\endgroup$
  • 2
    $\begingroup$ Are you able to share a sample dataset (perhaps the one in the above images) so we can try out some ideas? One thought to me is to use manifold learning to transform the data into a space that's more simple to work with and then perform the path segmenting like you are aiming for on the data in that new coordinate frame. $\endgroup$ – spektr Dec 26 '18 at 3:13
  • $\begingroup$ @spektr Thank you. I have prepared a sample dataset. Here is the link: 1drv.ms/u/s!ArdKO5IJqtSdgfdb9-q6qbym_tasnQ. It is shared via OneDrive. I will also read about manifold learning. $\endgroup$ – Anthony Dec 27 '18 at 2:15
  • $\begingroup$ @spektr I had a quick read about manifold learning. It seems to be a method for data with multi dimensions. However, my data have, in fact, only 2 dimension. The colours can be ignored as they only indicate the amplitude of the ultrasonic signal of the data point. I have tried to use the amplitudes to identify certain pattern but it is not very reliable, mostly because amplitudes are not consistent. $\endgroup$ – Anthony Dec 27 '18 at 10:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.