3
$\begingroup$

I have been reading the paper entitled Measures of Similarity by Unnikrishnan and Herbet, in which they discuss some metrics to evaluate image segmentation. They use the word refinement of labels/segmentations/images in different places. However, it is unclear to me what this means, and an online search for "image refinement" does not bring much enlightenment.

I hope someone can elaborate a bit on the meaning of refinement in the context of images, or more specifically if possible, in the context of image segmentation.

Thank you in advance!

$\endgroup$
3
$\begingroup$

The correct context of the refinement key word is segmentation.

Label Refinement in the context of image segmentation is a step to increase the resolution and understanding of the segmentation.

It can be done by exterior knowledge (Like labels on features) or other optimization steps to have a better results of the segmentation (Which basically labeling of the image pixels).

For instance, if the segmentation is based on Super Pixels a Refinement Step could be merging of some Super Pixels.

You may have a look at Label Refinement Network for Coarse to Fine Semantic Segmentation. In this paper they use Multi Scale approach for label segmentation. By using coarse level to refine finder level.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.