• Log in with Facebook Log in with Twitter
Photo

A little experiment (auto-distortion correction)


  • Please log in to reply
4 replies to this topic

#1 Klaus

Klaus

    Chief Editor

  • Moderators
  • 5,290 posts
  • LocationSydney, Australia

Posted 20 May 2017 - 03:54 AM

I took a perfect (artificially created) distortion chart.

Then I distorted it to 7% and corrected it back (manually - not 100% accurate but good enough).

 

I've attached a crop 700% enlarged that illustrates the effect.

The "perfect image" does not show any grey values - just black & white, of course.

You may see that one pixel row got pretty dark (loss of edge contrast). Or in other words - loss of resolution.

 

It's, of course, no surprise but it clearly shows how lossy a correction of a rather "typical" MFT wide angle distortion really is there.

Attached Files

  • Attached File  auto.jpg   211.48KB   5 downloads

Chief Editor
photozone.de

#2 JoJu

JoJu

    Advanced Member

  • Members
  • PipPipPip
  • 2,273 posts
  • LocationSwitzerland

Posted 20 May 2017 - 07:58 AM

I think, the lens distortions and their corrections make a worse impact than your experiment shows.

 

the corrections have to increase at the corners, right? And where are lenses showing the worst performances? Exactly these bloody corners. By interpolating pixels which are simply not existing and doing so for an average distance, I better look to cover all corners with bokeh  :D . 


  • Brightcolours and obican like this

#3 Brightcolours

Brightcolours

    Advanced Member

  • Members
  • PipPipPip
  • 4,562 posts
  • LocationThe Netherlands

Posted 20 May 2017 - 08:45 AM


 

I took a perfect (artificially created) distortion chart.

Then I distorted it to 7% and corrected it back (manually - not 100% accurate but good enough).

 

I've attached a crop 700% enlarged that illustrates the effect.

The "perfect image" does not show any grey values - just black & white, of course.

You may see that one pixel row got pretty dark (loss of edge contrast). Or in other words - loss of resolution.

 

It's, of course, no surprise but it clearly shows how lossy a correction of a rather "typical" MFT wide angle distortion really is there.

How did you distort it to what your test software calls "7%"?

 

Photoshop uses a different metric.

 

You have to make a distorted chart from the outset, not make one, then distort it (because the distortion will lose resolution, if you do that). 

 

This is what I end up with. I made a 6000 x 4000 image in PS. I put a chart in the top to "eyeball" about 6% using the Oly 12-100mm f4 PZ chart to judge the curvature.

 

I then pasted a pattern into the corner, of which you can see a part in the crops I made. I "undistorted" the chart again, and this is what the corner looks like after that. 

Left is the pattern I pasted onto the distorted chart, right is the same corner of the 6000 x 4000 image after correcting for barrel distortion.

The shift you see in the pattern illustrates pretty well the actual resolution that gets lost.

Attached File  distortiontest1.jpg   35.66KB   0 downloads

400% :

Attached File  distortiontest1a.jpg   68.21KB   0 downloads

 

Any tips on how to do this more accurately concerning what the test software calls "7%" are welcome.



#4 Klaus

Klaus

    Chief Editor

  • Moderators
  • 5,290 posts
  • LocationSydney, Australia

Posted 20 May 2017 - 09:51 AM

Well, I only did a quick and dirty distortion via the PS RAW plugin - thus I raped the lens distortion slider in there.

Then I measured the distorted image which turned out to be 7%.

No magic :-)


Chief Editor
photozone.de

#5 Brightcolours

Brightcolours

    Advanced Member

  • Members
  • PipPipPip
  • 4,562 posts
  • LocationThe Netherlands

Posted 20 May 2017 - 09:53 AM

Well, I only did a quick and dirty distortion via the PS RAW plugin - thus I raped the lens distortion slider in there.

Then I measured the distorted image which turned out to be 7%.

No magic :-)

Thanks for the explanation :)






0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users



© by photozone.de