Degrading Image Quality By Rotating Images

<p>Is it true that rotating an image in Windows Picture Viewer and/or Photoshop reduces image quality? And is it significant or is it some neurotic idea that I don't have to worry about. If it is significant what is the solution?</p>

I've wondered about this myself from time to time. I have always figured it would lead to some degradation, although whether it would be at all noticeable i'm not sure. The reason for it to degrade would probably be because at some point in the image, rotation would lead to say a multiple of a whole pixel (e.g. 1,2,3,4,5 etc. pixels) movement, while at other points in the image the movement would be in non-whole pixel multiples (e.g. 1.2, 3.5, 4.7 etc). These non-whole pixel movements would have to lead to degradation/interpolation.

 

I guess the best way to know would be to do a test. Take an image, duplicate, rotate the copy by a couple of degrees, then rotate it back again to the original orientation. Compare the copy to the original image using the difference (or is it subtraction?) layer blending mode.

<p>In my experience the reduction in image quality from techniques as rotation (say to level a horizon) or correction of barrel/pincushion or reasonable amounts of perspective distortion is generally completely insignificant even in a good sized print.</p>

<p>I have no dout that there is <em>some</em> degradation. But that is not really the core question. The important questions are a) is the distortion visible in the final print or other version of your photograph?, and b) would the uncorrected version, while less "degraded" from correction be better despite the remaining uncorrected tilt, etc?</p>

<p>Dan</p>

<p>I was thinking more when rotating an image 90 degrees, such as when taking a vertical oriented picture and then having to rotate it 90 degrees to view it properly on a monitor. I can't notice any difference at all when I do this and thought that there might be something going on that I'm not looking at or seeing. Sometimes I read these forums and say to myself 'you've got to be kidding' at some of the minutia that gets brought up. But then on the other hand there might be something to it...</p>

<p>I think that if your original is jpeg and you then rotate and re-save it in the rotated form, you will have degradation, but I don't know how much.</p>

<p>I just tried it, rotating a 16bit tiff 90 or 180 in Photoshop CS3 produces a file size identical to the original. Comparing the two at 100% I cannot see any difference. I frequently rotate a small amount to correct horizons etc and have never noticed any degradation in quality, but I will be looking a bit closer now ;-)</p>

<p>With jpegs some degradation would occur each time the file is saved, regardless of what editing operation you do.</p>

Many JPEG viewers and editors, including Irfanview and GIMP, offer lossless JPEG rotation. If the picture dimensions are not divisible by 8 you lose some from the edge, but with color images never more than 7 pixels.

<p>To the respondents so far, I would like to point you to John emphasising that he is talking about 90 degree rotations. In our bulk film scanning business we return all work to clients "correctly" rotated, so I have thought about this when is comes to doing post scanning rotation in 90 degree increments. We do all this sort of stuff it in Photoshop, as it seems to be the best way to finely manage pixels. Also, we do the rotations while in TIFF format, before we generate the JPEG files.<br>

I might do some tests on JPEG rotations, in Windows, Mac Preview and Mac Photoshop.<br>

Otto at Oscans</p>

 

<p>90-degree rotations of jpeg images can be done completely losslessly. If you use a specific tool like jpegtran you're certain not to lose any information at all. Most serious tools like Photoshop or Gimp are pretty certain to do this as well, though I suspect that if you do anything else (crop the image for instance, or edit it in any way) it will no longer do it losslessly.</p>

<p>That said, I did some experiments with jpeg encoding for a work-related project. At high quality (say, quality setting 95-99) you can unpack, edit and reencode the image many times without getting any visible degradation whatsoever. It is a lossy format of course, but after that experiment I feel that people are probably more worried about this than really is warranted. I know I was until I actually tried it.</p>

<p>Insignificant. There are other MUCH more important things to worry about...</p>

<blockquote>

<p>I was thinking more when rotating an image 90 degrees, such as when taking a vertical oriented picture and then having to rotate it 90 degrees to view it properly on a monitor.</p>

</blockquote>

<p>No degradation in that example (or 180), JPEG not withstanding. </p>

<p>To test these kinds of things, forget using your eyes <g>, instead try this:</p>

<p>

<p >To test a scanner for registration, scan the image. Change nothing, and scan the image again. For Raw processing or other kinds of tests, make two files of the same pixel dimensions (they must be identical. </p>

<p > </p>

<p > </p>

<p >In Photoshop, open both scans. </p>

<p > </p>

<p >Hold down the Option or Alt key, and go to Image > Apply Image.</p>

<p > </p>

<p >Set whichever image isn't listed as the target as the source. Set the Channel as RGB. Set the Blending to Subtract, with an Opacity of 100, a Scale of 1, and an Offset of 128.</p>

<p > </p>

<p >If the scans/test images were truly identical, every pixel in the image would be a solid level 128 gray. Pixels that aren't level 128 gray are different by the amount they depart from 128 gray. You can use Levels to exaggerate the difference, which makes patterns easier to see. Random speckles are noise, embossed effects are errors in the zero position, or misregistration -- if there's color fringing along with the embossing it's probably misregistration. Single noisy pixels show up as vertical stripes. Lamp variation as horizontal bands.</p>

</p>

<p>With tiff it's my understanding there will be no degradation doing a right angle (ie, 90 deg, 180 deg, mirror) transformation. By definition I think there would be some rearranging of pixels in non-right angle rotates, but I don't think it would introduce artifacts.</p>

<p>With jpeg, I think right angle tranformations are also nondegrading, at least that's what ACDSee tells me as it's doing them. I think non-right angle transformations would degrade, tho.</p>

<p>Thanks to everyone for your comments. And thank you to Bernie and Rodney for a testing methodology which seemed semi-similiar but I used Rodney's. Here is what I did: I made a copy of a JPEG file, rotated it 90 degrees in Photoshop, saved it, rotated it back 90 degrees and saved it again. Then I used the Apply Image procedure which turned the image neutral grey. I then examined the file at Actual Pixels and at 500%. At first I thought I found something but then realized it was some dust on my monitor. Another was a single dead pixel that stayed stationary when I moved the image around. After panning over the image at the two sizes I found... NO DIFFERENCE!</p>

I do not believe Photoshop is capable of lossless JPEG rotation <a href="http://www.betterjpeg.com/jpeg-plug-in.htm">without a plug-in</a>. Perhaps Adobe added it since CS2? John, to compare objectively, you must use the Difference layer, not mere eyeballing.

<p>90 & 180 degree rotations are (can be) lossless.</p>

<p>Anything in between will not be lossless, as there'll be some interpolation involved for some pixels. However, I suspect that some software packages apply a low degree of sharpening to the image after rotations to counter the concomitant loss of sharpness with 'arbitrary' rotations (as Photoshop calls it).</p>

<p>Why? Because if you take an image and apply, say, a 13 degree rotation to it in Aperture, Lightroom, Photoshop, Picassa, etc., you will see that the same image, which appeared equal in sharpness across all programs prior to rotation, appears softer in, e.g., Aperture or Picassa as compared to Photoshop (at least the old version of Aperture where I tested this).</p>

<p>In the end, it's really not a big deal. Having a slanted horizon (if undesired) is a much bigger deal and, like I said, a small amount of sharpening fixes the problem.</p>

<p>Rishi</p>

<p>huh!? Re-saving jpegs degrades IQ.</p>

<p>The 90 degree rotation you are talking about DOES NOT modify the image. It simply changes the orientation indicator in the Exif data. There is no change to the photo itself. Therefore there is nothing to test.</p>

<p>I wonder if a mirror of a jpeg could be lossless?</p>

Yes Mendel, Irfanview has vertical- and horizontal-flip among its choices for lossless JPEG operations. Carol, it is not the EXIF data that gets modified, but the JPEG quantization tables. For more information, see Gordon Richardson's <a href="http://www.photo.net/learn/jpeg/">excellent article</a>.