X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 12 Jun 2017 08:51 AM by  MariM
background pixels in classification
 3 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:8
New Member


--
31 Mar 2017 05:25 AM
    Hi, I want to classify SPOT 7 multispectral image in ENVI 5.1. After I created subset image of my study area using ROI, background data (the pixel values outside of the ROI data) occurred. Background data has zero pixel value. In order to mask background pixels, I created ROI from band thresholds and chose thresholds parameters for all bands as min value=0; max values=0 and then saved ROI to .xml. I selected the subset image as classification input file. I imported "mask ROIs xml" using mask options. . I selected “Selected Areas "Off” option from the mask definition menu bar. After I created the mask file, I selected mask band in classification input file menu. Mask band has pixel values=1 for subset part and pixel values =0 for mask part. After classification process, background or masked pixels was assigned as unclassified class. I want to leave those pixels out of my classification, because unclassified class affected class statistics results and classification accuracy results. How can I do this? Thanks.

    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    31 Mar 2017 09:57 AM
    What classification tool are you using? I no longer have ENVI 5.1 installed for testing but when I use a mask band in the Classification workflow, the output image sets the masked areas as 'Masked Pixels' which is separate from the 'Unclassified' class. So in ENVI 5.4 and 5.3, these areas should be separate. Do you have access to a newer version of the software?

    AYSIN KUZUCU



    New Member


    Posts:8
    New Member


    --
    12 Jun 2017 01:48 AM
    In order to mask background pixels, I used "edit envi header". I set data ignore value as zero. And then I applied supervised classification method such as max. likelihood classification. After classification process, background or masked pixels was not assigned as unclassified class. A pixel class has been created that is "masked pixels" separately in the classified image. Therefore, unclassified class did not affect class statistics results and classification accuracy results.

    Thanks for suggestions.



    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    12 Jun 2017 08:51 AM
    This is intentional. Pixels within the image that allowed to be processed in the classifier can come out as 'unclassified' due to the training data selection and classification thresholds used. These pixels are different than background pixels that are masked which are left out of the analysis altogether. These will have a value of 'masked pixels'. If you do not mask them, I believe the background will come out as unclassified since it will not match any of the training data.
    You are not authorized to post a reply.