Grid Pattern in Images after Dynamic crop

[gvanhau]

Hello

I was generating a star mask and was getting a very strange grid pattern in the background of the resulting mask.
First I tried modifying the parameters to see if it had to do whit them, but no matter what parameters, the grid pattern was there...
Then It came into my mind that it could be the image that had this pattern hidden into it.

In order to find it out, I applied a pixelmath  to the image, consisting of a discrimination of pixels by level (if pix<Value then 0 else 1)
When the value was somewhat lower than the background level, the pattern appeared.

I then reversed one by one the process, and checked if PixelMath still generated the pattern, and so it was until the very first stage on processing:

After the initial integration, and the Dynamic Crop the pattern appeared.
Before the dynamic crop, the pattern was not there.

I attached the image after integration, the image applying the dynamic crop, The image after applying Pixel math and a star mask of the same image.

Edit:
I just tested the different interpolation algorithms, and in only happens using bilinear and Bicubic Spline.

Edit 2:
I now tested this also on an image of another camera with  sony chip and got same result  the posted images are with KAF8300 chip.



Regards
Geert

[gvanhau]

Hello
I carefully re checked this, and it really happens with all interpolation algorithms, except nearest neighborhood.
The intensity of the grid depends on the crop angle and the selected algorithm.
If angle is 0 (no rotation) there is no such effect.

Can someone explain what is happening?
I'm afraid that this is causing the blotchy background we are seeing during processing of many images.

Regards
Geert

[dnault42]

This is a common problem with all interpolation algorithms.  The PI team's own documentation on this explains it much better than I could.  Go to the 'Rotational Examples' section.  There are similar issues with interpolation algorithms when faced with small scale differences.

Personally, I try to avoid all interpolation after integrating an image.  I do a two pass registration where the first pass is used to create an image that is cropped and framed the way I want and is used as a reference frame for a second pass of integration (using the original calibrated data).  This way most of the interpolation artifacts are removed by the rejection and averaging that occurs when stacking data.

Regards,
David