If you observe the results of many image segmentation algorithms, you can notice the occurrence of region borders where a human observer wouldn't draw one, for instance across slow gradient/textured areas. This seems to be a universal phenomenon.

I have annotated the image below with yellow arrows to show what I am thinking about.

My question is twofold: is there a fundamental reason why these occur ? How can we detect such lines ?

enter image description here

  • $\begingroup$ I am not sure this forum is the perfect place, but I found nothing better. $\endgroup$ – Yves Daoust Sep 27 '17 at 9:12
  • $\begingroup$ The fundamental reason they occur is because the segmentation algorithm finds them. Why exactly depends on the algorithm used. $\endgroup$ – adrianN Sep 27 '17 at 10:51
  • $\begingroup$ @adrianN: why do so many algorithms produce ghost outlines ? $\endgroup$ – Yves Daoust Sep 27 '17 at 11:05
  • $\begingroup$ It may be far-fetched, but have you seen seam-carving for photos? If the seam is in the same place, the energy level is low in that curve. The segmentation algorithms loves regions, not half planes (with bumpy edge of course), so these may be artifacts, depressed regions, low-energy seams or artificial borders. Have you seen it in segmentation using region growth? $\endgroup$ – Evil Sep 27 '17 at 17:42
  • $\begingroup$ @Evil: yep, seam carving is a wonderful technology. My question is about ghost lines in segmentation. $\endgroup$ – Yves Daoust Sep 27 '17 at 17:46

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.