04-11-2019 02:16 PM
I am pulling my hair out trying to understand how to export a TIFF image from ERDAS where the band composition is fixed as RGB and cannot be misinterpreted in another software. I am testing a supervised classification system on Picterra and when I upload a 8bit unsigned 3 layer image it reads it incorrectly and puts the bands in the wrong order. I would like to fix the band composite please tell me how to do this!
04-11-2019 02:43 PM
First off - What band order is your data in? RGB or BGR? Or, indeed, does it have more than 3 bands?
Knowing that we can narrow things down. Oh, and what version of ERDAS IMAGINE?
I'm not familiar with Picterra - what band order does it expect a 3band image to be in? Band 1 = Red, Band 2 = Green, Band 3 = Blue I assume, but want to confirm.
04-12-2019 03:09 PM
Thanks for the assistance. The file is in RGB order (just 3 bands) and Picterra should be reading it as RGB but it is failing to do so as it seems to reorder the bands. I am using version 16.5 of ERDAS. I want to export so the bands are stacked together in a fixed format as a single geotiff so Picterra reads it in a correct order.
Your help is much appreciated.
04-12-2019 05:30 AM
When you run the Exporter you use options like this?:
If so then I'm not sure why Picterra wouldn't interpret the data as RGB being bands 1,2,3. The Photommetric Interpretation tag should be set to RGB (you can check using ImageInfo) and other software should be capable of using that to set the default display parameters.
Do you know what default LUT stretch Picterra applies when displaying data and is it the same as the stretch you're using in ERDAS IMAGINE (which by default should be a 2.5% / 1% percentage linear)?
You may want to contact your local support (Geospatial Insight), or even contact Picterra's support, to try to pin down what the issue is.
04-23-2019 05:55 AM
This is Julien from Picterra. I just found out about this discussion.
As Ian pointed out, we should interpret RGB geotiff currently: we use gdal to extract the photommetric interpretation tag and figure out which band is what.
If the data is RGB uint8, we display the image as-is. Otherwise, it undergoes what we call 'normalization' (including LUT stretch as Ian mentioned), which is explained in more details here :
So if this is not working, it is possible we have a bug in the way we handle geotiffs coming out of erdas. If that is the case, please reach out at email@example.com and we'll have a look at the particular file.
Lead Software & Machine Learning Engineer