CRF for Improving Remote Sensing Multiband Image Segmentation

[Conor Cahalane]

Hello,

I am new to pystruct but would like to use it to improve segmentation following a SLIC superpixel creation process in Sci-Kit image. As per the sci-kit user guide "These superpixels then serve as a basis for more sophisticated algorithms such as conditional random fields (CRF)." link

I have been looking for a good guide on how to do this, but most of the guides I find confuse me. I found one that I liked Linky but it is for language processing, and whereas I understand how CRFs help in this case, i.e. 'y' is most likely to follow an 'l' in a sentence, I do not see how this would apply to land cover and mapping. 'roads' arent always beside 'grass'. Is there some other concept I am missing here? Probably. I also do not get how much of the process is automated, I think all? Or is there a supervised element of it.

Another method regional adjacency graphs or RAGs, help to join adjacent superpixels that are similar - would these be of any use? Should the workflow be SLIC >> RAG >> CRF? 

Some of the terminology confuses me, so thanks in advance for any explanations.  

Conor.

[sinno...@gmail.com]

Hi,

I am also working on this domain so I just wanted to ask you that have you got the answer to your query?If yes then can you please illustrate how have you done this.