X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 09 Feb 2016 09:34 PM by  anon
Background data in confusion matrix
 3 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:
New Member


--
09 Feb 2016 09:34 PM
    While performing Confusion matrix, how much it effect the utilization of background data? What it means by the term background data?

    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    11 Feb 2016 06:16 AM
    By 'background data' do you mean the pixel values outside of the image or in the background (usually all 0 or No Data)? If so, I do not think that the confusion matrix will be affected by the results as long as the background is included in the Unclassified class. Often you would use a mask in these areas when performing classification so that the masked areas are Unclassified.

    Deleted User



    New Member


    Posts:
    New Member


    --
    11 Feb 2016 08:21 PM
    By "Background" i means those pixels that were not assigned any class... or more likely 'Unclassified'. While performing "Confusion matrix using ground truth image"... Should i incorporate those backgroundbackground as class or not?

    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    12 Feb 2016 01:36 PM
    I noticed that if you perform a supervised classification with a mask, the masked pixels are called 'masked pixels' and assigned a class value that is number_of_classes+1. So these pixels (what I think of as background) should be separate from pixels that remain unclassified during a classification due to the thresholds used or the training data. In this case, you do want to keep the 'background' or masked pixels separate from the 'unclassified' pixels which is a valid class that should be considered. So it depends on if you have actual 'unclassified' pixels. If unclassified, then yes, include them. If 'background' or masked, then I would not include them and they should not be combined with actual unclassified pixels within the image.
    You are not authorized to post a reply.