Friday, August 22, 2008

Lightroom artefacts even show up when cropping

I don't know what causes the artefacts I talked about earlier. I do know they show up when you resize images on export. They also show up if all you do is crop your image. I have generated an example of this, again using the car image from before. I once exported the image uncropped from Lightroom to sRGB tiff, and I cropped it and again exported it. I took the two files into Photoshop, scaled them both to 200% using nearest neighbor scaling and aligned them to each other using the image align built in. Here is the result again using the mouseover trick. The default shows the image that was not cropped in Lightroom. I also include a substracted image that shows just the difference between the two calculated using apply image. The images can take a while to load after you hover so be patient. The change will be VERY obvious.

Not cropped in Lightroom

Cropped in Lightroom

Difference image

You see that is pretty bad. You see it is not just an extra halo that gets created, but also a sort of combing effect arises. Again Lightroom did not scale upon output. This was just the result of cropping. You can see these artefacts immediately in the Develop module in Lightroom when you just cropped.

P.S. I am not the first to point out this problem. Many people have done so before me (example link). I hope this will soon get fixed. You need light objects on dark backgrounds to see these effects (or the inverse). You see that even in pretty standard sort of images these problems can show up. If your image is not so sharp to start with this (the crop-induced artefact) will probably not be an issue. This was shot with a very sharp prime lens that outresolves the D300. However the resizing artefacts will show up even for less sharp images.


  1. Interesting finding! I hope you get it resolved.

    could it be that light room is putting a sharpening component somewhere in between cropping and exporting?

  2. Yep! There's also halos when cropping, not just resizing. Except they are not.

    When I read the post, it got me thinking why the 'non-cropped' and 'cropped' samples do not really align properly? This led me to do some experiments and get following results.

    The resize artifacts do not always appear on a cropped image. The trick is that Lightroom crops in real coordinates, rather than integer. And if your crop rectangle happens to run through a row of pixels, rather than nicely in between rows, the exported image will have to be resized to the nearest integer size - you can't have fractional pixels, can you?

    So the bottom line is... even in the case of cropping artifacts, the image is actually RESIZED, though only by a fraction of a pixel.

    P.S. Given that, the 'difference' image test is not very valid, because you a comparing one image to a different one that is resampled (even if only by half a pixel) to match the first one. Hence the 'combing' effect, it's not really there, it's a result of comparing mismatched images.

  3. Indeed, the obvious explanation is that Lightroom also resizes besides just cropping. About the difference image, you can clearly see the combing in the cropped image in Lightroom, not just in the difference image. I think it is a real artifact. Certainly very annoying I think.

  4. I decided to post a topic about this issue on the ACR forum ( The most irritating that there's no such feedback on the forums as, say, from the 'Adobe reds' issue, which has, as we know, eventually been solved.

    There's been some random post on the Lightroom forum, regarding the related resize problem, but it has not generally caught the attention of public or the 'big guys'.

    There's much less traffic on ACR forum and apparently more attention from guys like Eric and Thomas there. So, maybe it helps.

  5. Hi

    I've come from the Adobe LR forums. Would cropping/resizing in the Genuine Fractals LR plugin solve this problem?



  6. Michael,

    I guess it should solve it indeed.

  7. I'm not fully buying the resize-argument, personally I think it's a bug in the subpixel interpolation code, but then again I don't know how they actually implemented the crop.

    If you would compare crops for each corner you could easily see if all corners shifted the same way (your example picture shifts vertically), or if they shifted "out from center" or "in towards center" respectively. If they shifted the same, it's the interpolation, if they shift differently it's a resize.

  8. I think you are right that the stair stepping clearly is a bug in the sub-pixel interpolation. However the halo, I think, is coming from the lanczos resize. It is the same as what happens on a resize and lanczos is known to do this when the parameters are not fully tuned. Anyway I have no inside knowledge here. I trust it will be fixed soon though.

  9. In fact I know it will be fixed soon.

  10. > If they shifted the same, it's the interpolation, if they shift differently it's a resize.

    It's the same thing in this case.