Topic: Correlation errors

I am trying to use statistical correlation to measure offsets between a pair of GF1 images. The GF1 satellite was launched by China and provide images with 8m spatial resolution. I have asked if the cosi-corr could be used to measure the displacement between the GF1 images, and it is said the cosi-corr could deal with GF1 images after they were orthorectified.
Now, I have orthorectified the GF1 image pair and made a small subset for my region of interest. But, when I started the correlation, I got two errors message saying "CORRELATION_IMG_CCALL --> Images coordinates origins must be a multiple of the resolution for a grided output." and "CORRELATION_IMG_CCALL --> Images connot be overlapped due to their origin and resolution - Origins difference greater than 1/1000 of the pixel resolution."

Please could anyone suggest what the problem is and how to sort this out?


Re: Correlation errors

can anybody have seen these errors and know why?

3 (edited by ehoffman 2016-05-29 22:25:26)

Re: Correlation errors

i tried to correlate Landsat5 images of the Path/Row combo:232/79, 233/79 und got very striped correlation results, for each single optical band 1-3 (i.e. DOY 2003340 - 2004359). i got the same bad results for any path/row-combo and timeperiod. The sources of the images is the L1T-preprocessed data from usgs earth explorer. (Spatial Resolution: 30m/Pix, DN:0-255)
I linked a sample image here:
http://www.directupload.net/file/d/4371 … n8_jpg.htm
What is the reason for the stripping?
Thanks for help.


Re: Correlation errors

Hi cmylovemumu,

you can solve the first error message by unchecking the "Gridded Output" option in the correlation setup window. Please refer to the User Guide for more details on that option

Regarding the second error message: In order to be compared the two orthorectified images must be on the same ground grid and have the same resolution. I assume that during your orthorectification process, you made sure to orthorectify the two images with the same output pixel resolution. You must also make sure that the ground grid of the two orthos can be aligned. For instance if your resolution is 10m/pix and the top left coordinates in EW of the first ortho is, for instance, 150000. Assume now that the top left coordinates in EW of the second ortho is 150112. Then, because the resolution is 10m/pix, you can see that you won't be able to align the grids perfectly. There will always be 2m of shift that will remain. If the top left coordinates were 150110 instead, then a shift of 11 pixel (i.e. 110m: 10m/pix x 10 pixels) then the grid would be aligned.
When setting up the orthorectification, you need to adjust the output grid so they can be aligned during the correlation process.



Re: Correlation errors

Hello Erik,

indeed this is a strong striped correlation results.
It would be surprising that it's an actual jitter of the platform. I would think that it's related to a bias in the measurement.
From the name of the file you sent, I presume that you did a correlation in Fourier with a 16x16 window size. Could you try 32x32 and see if the stripping is still there?


6 (edited by ehoffman 2016-07-12 00:19:56)

Re: Correlation errors

Hello Mr. Ayoub.
thanks for the reply. I have tried to correlate the original data with some Window/StepSize Combos (e.g. 32x32 and 64x64 windowSize) but got still the same stripping in the correlation.
For the tests i used an actual Landsat5-Dataset from P-R-Combo 232/79 between the years 2003 and 2004.
I have also tried an ENVI-Fourierfilter, but it failed to a black NAN-filter output (maybe because of the "FLOAT"-Value of for each pixel, but its only my speculation).
Do you have another idea for the reason of stripping or for a filtermethod?
If you want, i can send the original data to you.
Thanks fpr your help.


Re: Correlation errors

Hello Erik,
interesting. I haven't seen such stripping with Landsat before with adjacent row/path images. Could you tell me what is the amplitude of the jitter that you see?

8 (edited by ehoffman 2016-07-28 00:28:53)

Re: Correlation errors

Hello Mr. Ayoub,

for completion here are all relevant parameters of the correlation.

First Image: LT52320792003340CUB00_B2.TIF
After image: LT52320792004359CUB00_B2.TIF

Frequency, Ini.WindowSize: 32x32, FinalWinSize: 32x32, Step:3, Robustness:4, Freq Tresh: 0.9

For an area of constant jitter, the Average Amplitude of the jitter: ~13m and the wavelength: ~ 11m
Measured values come from a profile perpendicular to the strip-pattern (jitter)

(See linked picture below)

http://www.directupload.net/file/d/4429 … iw_jpg.htm

Are these values, that you need?


Re: Correlation errors

Hi Erik,

thanks for the information.
It looks like the "jitter" is in the images and not a correlation artefact. I haven't seen such jitter previously in Landsat imagery. Did you try correlating other bands than B2?  Also, you could try downloading and correlating another pair from the same path/row and see if you get similar jitter.
In any case if the jitter is in the Landsat, you have no other choice than removing them in post-process.



Re: Correlation errors

The jitter appears for different bands and different time-intervalls.  I also tried a correlation by subsetting, but it fails into striped output

I have uploaded the 2 original Landsat5 Scenes. In my opinion, there is no jitter in original two scenes, isn`t it ?!

2003: http://www.bilder-upload.eu/show.php?fi … 137880.jpg

Thank you.


Re: Correlation errors

Here is the Scene in 2004:

2004: http://www.bilder-upload.eu/show.php?fi … 138026.jpg