Perspective Correction: Lens vs. Software

I’ve been coming across a lot of discussion of tilt-shift lenses recently. One of the particular benefits of these lenses is the ability to apply perspective correction to a scene by shifting the lens up or down. This allows the photographer to keep the camera level, which means that vertical lines stay straight instead of leaning backward (if you’re pointing the camera up toward the sky) or forward (if you’re pointing the camera towards the ground).

I use the shift function extensively in architecture photography, where it’s especially important to keep lines straight. But, shift can be useful elsewhere too. In landscape photography, for instance, a photograph generally works better if the horizon is not in the center of the frame. But, if you compose with the camera level, that’s just where the horizon will be. To move the horizon off center, you can point your camera up or down. That may not big a big deal. But, let’s say you have some nice tall trees in the picture. Those trees will end up looking like they’re leaning. And, the wider the focal length, the more problematic this can be.

One way to solve this is to use a T/S lens to shift the composition. Another way is to use perspective correction in software after the fact. Now, some people will tell you that correcting perspective in software can cause a degradation in image quality because you’re pushing pixels around in your image. Technically that’s true. But, unless the corrections are extreme, the degradation is pretty small. Honestly, I wouldn’t be too worried about that unless you’re making large prints. However, you do need to be concerned about cropping away a lot of your composition.

Here’s an example. These images were shot with a Canon 17mm TS-E lens. The top image is unshifted, with the camera tilted up to capture the entire tree in the frame. The bottom image is with the camera level and the lens shifted up about 6mm. (Keep in mind that 6mm is what I’d consider to be a moderate amount of shift…full shift is 12mm.)

Now, let’s look at what happens to the unshifted image when we apply perspective correction in software to try and match the shifted image. The software has to progressively stretch the image out towards the top, while pulling it up at the bottom:

So once you’ve done this, you have to crop the image:

We can correct the perspective to make it look a lot like the shifted version, but we end up having to cut out a significant portion of the composition. In the unshifted version after correction, the composition ends up off balance with the tree now too close to the edge of the frame. Overall, in the corrected version, you end up throwing away about 20% of your original pixels.

As far as what the pixels look like themselves, let’s have a look at that slight degradation in image quality.

Unshifted, corrected in software:


So yes, correcting for perspective in software does cause a bit of degradation, but I don’t think it’s extreme and probably won’t be noticeable unless you’re making larger prints.

But again, remember that the above example is only a 6mm shift. The more perspective correction that’s required, the more problems you’ll run into trying to do that correction in software, in some cases running your subject completely out of the frame. As an example, consider this image:

This image was also shot with a Canon 17mm TS-E, but at full shift. It would’ve been very difficult to try and replicate this composition with a non-shift lens and maintain the entire building in the frame (needing to resort to an even wider focal length, or stitching of multiple images).

So, in summary: yes, software correction can be a very useful tool to correct perspective if you’re not working with a shift lens. But, if you intend to use it, you need to be very careful when you’re shooting and plan ahead by incorporating extra space into your composition which will need to get cropped away.

This entry was posted in Equipment, Techniques. Bookmark the permalink.

Comments are closed.