1

I've processed a raster layer with pixel values ranging between 1 & 8 (see here). A cursory glance reveals that, there's a good tendency of clustering in different parts of the heterogeneous raster band, occasionally intruded by unwanted unlike pixel values. My goal is to make spatial clusters with the following characteristics-

  1. Every spatial cluster will represent only one pixel value i.e. all the minor intruding pixel values will be converted to the prominent pixel value surrounding it;
  2. There can be more than one cluster with the same pixel value, but has to be spatially well-separated;
  3. The clusters can be of any size &/or shape.

Here you can see a few clustering example zones drawn by hand to show tentatively how I want them to be classified. I have searched the 'raster' package of R, but failed to discover anything relevant. A painstaking solution to such problems is digitizing, which, in many ways, is unprofessional, especially when someone has to deal with similar situations over and over again.

Improve this question
3
  • 1
    Tricky, with lots of possible tuning parameters and subjective decisions to make, so will depend very much on your data and how well you feel any algorithm is doing. In your example, the lowest-rightest cluster looks like it should be two clusters - a cyan one and a green one, for example. Why have you defined it as one cluster?
    – Spacedman
    Commented Dec 29, 2013 at 12:54
  • You are correct @Spacedman, I should have defined it as two separate clusters while digitizing - all these are tentative, however.
    – ToNoY
    Commented Dec 29, 2013 at 17:41
  • 1
    I've had a few thoughts. First, simplify to two colours, A and B say, and think about that. I reckon you need to trawl the image analysis literature.
    – Spacedman
    Commented Jan 4, 2014 at 20:55

0

Reset to default

Browse other questions tagged or ask your own question.