0

Image classification

Solomon Gbanie

July 05, 2012 05:23

None

Follow

Dear All,

I am doing image classification and I am doing it with Segmentation module. Is the a minimun recommended tolerance?

• Facebook
• Twitter
• LinkedIn

5 comments

• 0
  
  Clark Labs Tech Support Team July 09, 2012 15:41

  Dear Solomon,

  The level of tolerance that you should specify in the Segmentation module depends on your data and your expected, or desired, result.  The minimum tolerance that you can enter is 0, which results in very small segments.  A larger tolerance results in larger segments (e.g. fewer total segments) and greater generalization.  The best way to find the ideal tolerance for your data is to test a few different tolerance values and compare the results to your expected outcome.  Generally, it is recommended that you start with a tolerance of around 10 and then modify it to get the best results.  There is no recommended minimum tolerance.

   

  David

  Clark Labs Tech Support

   

   

• 0
  
  Solomon Gbanie July 15, 2012 01:51

  Thanks for the response. I am doing some segmentation analysis. My Tiff image have a back ground value of zero. After attempting the segmentation of the classification which will allow me to do Segtraining, the result I got assign vales to the background which is outside my AOI. This will affect my result and I want to fix that up before proceeding with the analysis. Have you had experience of that type? What do you suggest to do so that after the segmentation part of the work, I will still maintain zero as my background. The image I am dealing with is a landsat image and it is not a square.

   

  Thanks.

• 0
  
  Clark Labs Tech Support Team July 16, 2012 20:02

  Hi Solomon,

  The background values should not matter because after running the SEGTRAIN module, you are only identifying the segments that you need for the analysis. 

  If you are still concerned, when you use the SEGMENTATION module (as you should before using SEGTRAIN), you can specify a background value and the flag definition field as “background” in the input image layer’s documentation (.rdc) file with IDRISI Explorer.

   

  Laura

  Clark Labs Tech Support

• 0
  
  Solomon Gbanie July 19, 2012 02:57

  Thanks for the information. I am however to fogure out how this works. Just a reminder, My tiff image has all a bacl ground value of 0. I develop a false composite image of ban 2_3_4. I then set out to do segmentation of all the six bands. Outside my AOI which initially have a zero value now assumes to be a  polygon and three polygons now exist outside mt AOI. When I went forther to do the classification with the segmentation. this three polygond which fall outside my study happen to be water after my classification. This has slow my classification as I wamt to be sure at the earlier stage before proceeding with the classification.

   

  Also when do the SegTran you are to overllay the vector file over the composite image. Since this is a vector file, is there a way to clip a vector in IDRISI so that whnw do the overlay for the SegTrain this poly area (three) will be out. If you dont mind, I can send you the files for you to beter understand what I am saying. it would be nice if I know what I am doing and to know what to do it.

  Thanks for your anticipated help.

• 0
  
  Solomon Gbanie July 20, 2012 02:40

  I have got it right. Thanks Laura