of the difference between post-processing perspective correction and in-camera correction using a shift lens. Because of the very wide angle lens used here, there is noticeable vertical foreshortening - in the same way that objects get smaller in a landscape as they get further from the camera, the features of the building get "compressed" as you look up the building. So, in this case, when the top of the picture is stretched to correct the verticals, the picture ends up looking "squished" vertically. Compared to the original, the people standing by the door and other features on the building look too wide in relation to their height. A more natural correction could have been achieved by stretching the top and compressing the bottom around the vertical mid-point of the image. The vertical foreshortening, while still there, would be less noticeable.
An even better example is to try this on a building which has equal-height windows running up it. With a shift lens and vertical film plane, the windows will all be the same height measured along any vertical line in the image - this is how the projection from vertical planes on the building to the vertical film plane works. Doing the correction in software by stretching the horizontals only will result in parallel verticals, but the height of the windows will decrease towards the top of the image. To get the same perspective correction as the shift lens, the image would also need a "progressivly" stretched vertically, that is, no stretch at the bottom with a linearly increasing amount of stretch leading to the top. This should be possible in software, but I'm not aware whether anyone's done it. It would, of course have the disadvantage of having to either interpolating or downsampling pixels to get the "correct" perspective.
The attached remake shows 3 examples of correction to this image. The first is the Shun Cheung's original corrections, the second uses photoshop's perspective tool to stretch at the top and compress equally at the bottom, and the third adds a bit of vertical stretch, which I think improves the proportions of the people by the door and the front of the limo (ok, it is a "stretch" limo). In photos taken at generally less extreme angle upwards and longer focal lengths, a gentle correction will usually not result in such noticeable vertical squishing. Sorry for the long-winded lecture, and please feel free to "correct" me on any of the points I've made!
Mission Santa Clara
in Uncategorized
Posted
of the difference between post-processing perspective correction and in-camera correction using a shift lens. Because of the very wide angle lens used here, there is noticeable vertical foreshortening - in the same way that objects get smaller in a landscape as they get further from the camera, the features of the building get "compressed" as you look up the building. So, in this case, when the top of the picture is stretched to correct the verticals, the picture ends up looking "squished" vertically. Compared to the original, the people standing by the door and other features on the building look too wide in relation to their height. A more natural correction could have been achieved by stretching the top and compressing the bottom around the vertical mid-point of the image. The vertical foreshortening, while still there, would be less noticeable.
An even better example is to try this on a building which has equal-height windows running up it. With a shift lens and vertical film plane, the windows will all be the same height measured along any vertical line in the image - this is how the projection from vertical planes on the building to the vertical film plane works. Doing the correction in software by stretching the horizontals only will result in parallel verticals, but the height of the windows will decrease towards the top of the image. To get the same perspective correction as the shift lens, the image would also need a "progressivly" stretched vertically, that is, no stretch at the bottom with a linearly increasing amount of stretch leading to the top. This should be possible in software, but I'm not aware whether anyone's done it. It would, of course have the disadvantage of having to either interpolating or downsampling pixels to get the "correct" perspective.
The attached remake shows 3 examples of correction to this image. The first is the Shun Cheung's original corrections, the second uses photoshop's perspective tool to stretch at the top and compress equally at the bottom, and the third adds a bit of vertical stretch, which I think improves the proportions of the people by the door and the front of the limo (ok, it is a "stretch" limo). In photos taken at generally less extreme angle upwards and longer focal lengths, a gentle correction will usually not result in such noticeable vertical squishing. Sorry for the long-winded lecture, and please feel free to "correct" me on any of the points I've made!
Jim