02-22-2018 03:38 PM
Dear Support team,
When I pan-sharpended an image which was subseted by aoi file (my study area), the result was showed as my study area filled with black color and the rest of the part of the image is white (see image below).
I have used Erdas Imagin 2016 and the satellite image was World View-2. Could you suggest me where the problem arises from? In addition, when I open the subseted raster image (my study area) which was taken from the orthorectification image in the 2D viewer, the name of a file on the top of the viewer converted in to "UNKNOWN-"name of file"-Country:N/A, Date:.........." (see image below).
Is that becasue of projection? Thank you so much!
Solved! Go to Solution.
02-22-2018 01:11 AM
First - read what data really is. Do not trust what you see, so right click - inquire cursor and see is there really pixel values in black areas or not. Your data looks a bit like case where background white (255, 255, 255) messes up the statistics and true data is so close to zero (or pure zero) and it gets lost in black.
So always step number 1 - try to figure out what data really is - do not get stuck on situation how it looks.
Maybe even more important tool in Imagine instead of viewer is Image Metadata and Inquire Cursor tools that give you access to real data. Viewer is just how it looks but data is something else.
02-22-2018 04:01 AM - edited 02-22-2018 04:04 AM
Thanks so much for your constructive suggestion!
I have tried to observe the result using inquire cursor. As you said the pixel values in the black area are more close to zero than the white area.
However, I need to see the real image (visible) since the purpose of this output is for classification. How can I use this black and white output to calssify? The two subseted images (panchromatic and multispectral) to pansharpend was look like the following (images below):
02-22-2018 04:20 AM
If you use pure calculative methods for classification then actually black&white does not matter. What needs to be done to get data better visible to you is get rid of that white and there are few methods:
1) Subset your data using polygon that follows your true data
2) Use spatial modeller to recode data so that all white (255,255,255) is turned to black (0,0,0)
3) Recompute statistics of your data in a way that you define NODATA value and Ignore stats value to be white you have there (=255 if 8 bit data)
Normal Recode tools of Imagine will not work as those are designed for thematic data and your data is most likely no thematic. The thing you see here is lack of image processing capabilities of some third party tool. For example for older ArcGIS it was very typical that data was build like yours to make background look nice in ESRI white background. If this is done in stupid way to actual pixel values you face issues like this. The proper approach is that bacground can be any value and it is then told in software what pixel value represents background/no data.
You probably can see something in your data if you just simply open file and while opening you state that "no stretch" is done.
What I do not know based on this talk is
a) Is your data really 8 bit as I refer or somethng different like float or 16 bit. That affects on pixel values and I talk about 8 bits
b) Do we have some kind of pan sharpening bug that generates data in wrong way. That you must analyze by looking the input data you feed. My guess is that another of them has something weird in background values that generates issue like this.
So hard to say without having the whole data in hands but things like this definetly involves here.
02-22-2018 12:16 PM
@tikola Thanks again!
My data before pan sharpening were unsigned 16 bit for both gray scale and multispectral (according the info in pan sharpening window) and I have not used any third party tool other than Erdas. I guess this may be from the version of Erdas (new feature). One person that I know, who has used 2014 version for the same type of data, has got a clear (visible) result of pan sahrpening. I will try to do every steps again if the problem can be solved. If you get any new solution, let me know. Thanks so much!
02-22-2018 01:09 PM
Can you please specify which Pan Sharpening technique you are attempting to use and what parameters you set on the dilaog?
02-23-2018 02:49 AM - edited 02-23-2018 02:51 AM
HCS Resolution Merge technique were used. And the parameters set in the dialogue looks like the following (see image below):
02-23-2018 05:52 AM - edited 02-23-2018 05:59 AM
My suspicion is that because the Hyperspherical Color Space technique involves a color-space transform it is getting thrown off by the predominance of n-dimensional "white" in your input image.
To test this theory try running your data through one of the techniques which does not rely on a color transform. HPF is probably a good candidate.
If that does turn out to be the issue, and you still want to stick with the HCS technique, I think you'll need to follow Timo's suggestion - recode the background values to 0s first. A simple Spatial Model should do that for you.
Oh, hang on. Your input data doesnt have white background. What are the background values (the "black" areas) in the input images (in all 8 bands of the MS and in the Pan)?
02-23-2018 11:06 AM
Oh - I think Timo was on the right track. All the areas that were 0s in the input images may have been mapped to 65535 (or similar) in the pan sharpened output. Those values will be swamping the statistics and therefroe the display LUT.
Open the pan sharpened image in ImageInfo. What's the Max value in each band? Something like 65535? If so go ahead and set that value to be the value-based NoData value. Then recalculate statistics (direct binning, Skip of 1, preferred).
Redisplay the pan sharpened image. Does it look better?