Luminous Landscape Forum
Raw & Post Processing, Printing => Digital Image Processing => Topic started by: Rich on May 28, 2007, 06:37:11 am
-
A lot of people moan about the price of Photoshop. I bought version 7.0 when it came out because it was a lot cheaper than buying a perspective control (PC) lens! The full version of Photoshop gave you the transform>perspective option to correct converging verticals. If you photograph a lot of buildings it is a very important feature to have. Photoshop is also a far better way to correct verticals than a PC lens (or using camera movements) because you are not using the edge of the lens with its associated vignetting and image softening.
I now have CS2. As far as I can see there are now 4 ways to 'correct' perspective. They are Edit>Transform>Perspective using top or bottom handles, the new (to CS2) Filter>Distort>Lens Correction and lastly using the Crop tool, and by ticking the Perspective option, Photoshop crops & 'corrects' the Perspective at the same time.
I have never seen in a book or magazine the correct method/tool to use, because if you don't use the correct tool in the right way, a lot of distortion results. I did a test to see which was best. I shot some graph paper on the wall with the camera looking up at it. I corrected the shot for barrel distortion (which now seems to be a standard feature with modern cameras).
Firstly, I corrected the Perspective by pushing in the lower handles using Edit>Transform>Perspective
The result is a perfect solution all the squares are square. This is the correct method to use. It maintains true geometry of the building.
If you pull out with the top handles using Edit>Transform>Perspective the verticals are corrected, but the image is distorted badly in the X direction.
If you use the Filter>Distort>Lens Correction, the result is good but still results in some stretching in the X direction. Ironically it is the toll I use the most because it's the nicest to use!
The worst tool of all is the Crop Tool. This results in big X distortion.
So now you know. At least you know you building is still the correct shape. I think Adobe should aim at making all the tools distortion free or issue a warning message for architectural Photographers.
-
Thanks for this interesting comparison.
Although I understand your feeling on this issue, a tool is a separate entity from the knowledge and skills required to use this tool to carry out a given task.
Trying to build usage knowledge inside the tool for a sub-set of users is generally speaking not a healthy direction to push the enveloppe of product capabilities.
You wouldn't think of writing on all the hammers that they can be mis-used in smashing one's fingers, would you?
Cheers,
Bernard
-
This issue also points towards the fact that Photoshop is not purely for photographers. If the program was we would have one perspective correction tool and that would not distort. From a photographic point of view, some users will assume tool ‘a’ does this and tool ‘b’ does that, because it says so, without realising there work is being distorted. It is the responsibility of the software writers to ensure the user automatically produces the correct result. If creative effects are required the user can always use a free form tool such as edit>transform>distort.
The ‘Lens correction’ tool I think is very much aimed at photographers although ironically falls under the ‘filter>distort’ pull down! This tool does not maintain true geometry as I feel it should.
One thing I have noticed when using ‘transform>perspective’ is that on a very tall building, the true result can look uncomfortably like the building is too tall and stretched. This is because we are used to looking up at tall buildings and not seeing them face on. Maybe this is the thinking behind the Lens Correction tool, because that tool will soften this effect, I don’t know.
-
This issue also points towards the fact that Photoshop is not purely for photographers. If the program was we would have one perspective correction tool and that would not distort. From a photographic point of view, some users will assume tool ‘a’ does this and tool ‘b’ does that, because it says so, without realising there work is being distorted. It is the responsibility of the software writers to ensure the user automatically produces the correct result. If creative effects are required the user can always use a free form tool such as edit>transform>distort.
The ‘Lens correction’ tool I think is very much aimed at photographers although ironically falls under the ‘filter>distort’ pull down! This tool does not maintain true geometry as I feel it should.
One thing I have noticed when using ‘transform>perspective’ is that on a very tall building, the true result can look uncomfortably like the building is too tall and stretched. This is because we are used to looking up at tall buildings and not seeing them face on. Maybe this is the thinking behind the Lens Correction tool, because that tool will soften this effect, I don’t know.
[a href=\"index.php?act=findpost&pid=120086\"][{POST_SNAPBACK}][/a]
-
Many thanks for the comparison-
(to start you turn the vertical axis to vertical before doing the perspective distortion you might add.)
Now for the horizontal perspective I believe that the standard tool don't work at all because it all depend on the height of the horizon in the picture- _ so i use distortion & my feeling to get it right - but there must be a more scientific approach...?
who?
So I am also looking for the right solution for these basic problems. In a way I feel that you can never do it right for we are transforming 3D to 2D...
kers
-
...One thing I have noticed when using ‘transform>perspective’ is that on a very tall building, the true result can look uncomfortably like the building is too tall and stretched...[a href=\"index.php?act=findpost&pid=120086\"][{POST_SNAPBACK}][/a]
Any time you adjust the perspective a significant amount it causes some stretching. It is wise to then strech out the image a bit in the opposite direction to make it look correct.
-
I'm misunderstanding something here re: stretching. I use the transform/perspective (or distort) quite a bit. When I correct for converging verticals it seems that the building shrink, not stretch. Sometimes I'll actually go back an rescale the image up the vertical axis a bit.
What am I missing?
-
I'm misunderstanding something here re: stretching. I use the transform/perspective (or distort) quite a bit. When I correct for converging verticals it seems that the building shrink, not stretch. Sometimes I'll actually go back an rescale the image up the vertical axis a bit.
What am I missing?
[a href=\"index.php?act=findpost&pid=200729\"][{POST_SNAPBACK}][/a]
You are right, that is actually what I meant. It probably is most common to pull out the narrow side making a building more squat looking.
I guess it depends on how you go about it. You can pull out the narrow side or squeeze in the wide side to get rid of converging lines. In any case you'll have to compensate by stretching or squeezing in the other direction to compensate or else the building will either look squat or stretched. (Of course it can be something other than a building.)
-
Interesting thread.
Does anyone have any thoughts about at what stage of the digital workflow it's best to carry out perspective / lens distortion correction? ie Before or after other retouching & colour / curves corrections etc.
Graeme
-
Interesting thread.
Does anyone have any thoughts about at what stage of the digital workflow it's best to carry out perspective / lens distortion correction? ie Before or after other retouching & colour / curves corrections etc.
Graeme
[a href=\"index.php?act=findpost&pid=200883\"][{POST_SNAPBACK}][/a]
If CA is present in the image, it will be made worse by any geometric alterations, sharpening, etc., so you want to get rid of that right away. Ditto on vignetting. I would definitely do lens correction before perspective. If you have a lot of barrel distortion for example, it will be difficult to evaluate the perspective accurately.
Here's my order of operations when using Lightroom/Photoshop:
1. Lightroom
a. Assign a camera calibration profile, if not handled automatically.
b. Adjust the white balance.
c. Use the tone controls to get the white point, black point and gamma close to final values. At this point, I err on the side of eliminating any clipping.
d. Level the image if needed.
e. Adjust the CA correction (Easier to do if the white balance is already reasonable; steps e., f. and g. should be done at high magnification).
f. Adjust the noise reduction controls (If the noise is really serious, skip this step in favor of more powerful NR tools in Photoshop. If this is the case, be very careful with the next sharpening step to not make the noise even worse).
g. Adjust the sharpness controls along "capture sharpening" guidelines.
h. Adjust the presence controls.
i. Re-evalute color and tone and adjust as necessary.
2. Photoshop
a. Immediately after opening, use NR tools if needed. Consider using masks or layers to confine heavy NR to appropriate areas, such as skies.
b. Use PTlens plugin to remove lens distortion.
c. Perform selection type adjustments (local color corrections, etc.), and curves adjustments. Depending on what I'm tryng to accomplish, I might switch to Lab color at this point.
d. Adjust perspective.
e. Crop.
f. Perform "creative sharpening."
g. Look at color and tone again.
h. Perform output sharpening.
Consider that everything in that workflow, except for 2.c., 2.f. and 2.h., can be done in DXO optics on the raw file, nondestructively. For example, you can go back and alter the noise reduction settings if it looks like there will be too much blurring. The lens correction step, including CA correction, is completely automatic, and requires no user intervention. If Lightroom could do really good lens correction (distortion, not just CA and vignetting), I might not need DXO. I think what the world really needs is a DXO plugin for Lightroom.
-
Interesting thread.
Does anyone have any thoughts about at what stage of the digital workflow it's best to carry out perspective / lens distortion correction? ie Before or after other retouching & colour / curves corrections etc.
Graeme
[a href=\"index.php?act=findpost&pid=200883\"][{POST_SNAPBACK}][/a]
DXO lets one adjust the perspective at the raw stage. At this point it will also correct for c/a vignetting, etc. However, if I have to combine parts of two images with different exposures, I do not make any perspective changes until the two images have been combined. I haven't seen any problem adjusting the perspective later rather than earlier.
-
It looks to me like we're talking about a couple of issues here, so I'll jump in on both of them.
Forcing that skyscraper into a rectangle when it's shot from street level can indeed look like distortion. It might be "right" but it doesn't agree with our mental perception of what a tall building seen from that point of view should look like.
Of course, the "falling backwards" building is a problem, too. That's why we go through all this perspective correction stuff in the first place.
So what's the solution? I don't think there is one, except to compromise. We live with a fundamental disconnect between real-life vision and 2-D representation. Nothing new there. The good news is that maybe we don't need to buy that 6mm shift lens after all. If you're shooting the AT&T building from the sidewalk across the street, relax. It ain't going to "fix". Even if it did, you probably wouldn't like it.
But if you have a clear shot at that building from a few blocks away, maybe it would be worthwhile to apply perspective correction. Go to it. But remember that getting that box into a perfect rectangle might lead it to look like it's falling forward. Whether our brain tells us in so many words or not, it will be happier if AT&T tapers toward the top just a little bit.
Okay, to blather on to issue 2. My own method for perspective correction, when I do need to get those edges good and parallel, is to stick with the old faithful Edit>>Transform>>Perspective. After getting the edges parallel, but before committing the change, hit CTRL-T (Free Transform), then hold the Alt (or Option) key while dragging out the boxes on the sides of the frame to the original frame edge. (Hold Alt/Option so the two sides move out symmetrically.)
My geometry skills are not strong enough to tell with confidence that my method "keeps the squares square", but I notice that the Lens Correction Filter also appears to rotate around the center line of the original photo.
Thanks for reading this far.
MB
-
Rich, thank you for posting the results of your tests - inspired use of graph paper, very interesting.
When I looked into this a couple of years ago, the common wisdom (on internet forums) was that when using Photoshop's transform perspective tool (to correct converging verticals), it was best to pull out the top corners to the same degree that you push in the bottom corners.
I processed a whole project using this method, but I always felt that the buildings looked slightly squat. Your test bears this out. I will reprocess, just using the bottom corners.
Cheers
Elliot