Skip to main content

Linear gradient after lens distortion correction

Comments

17 comments

  • John Friend
    Top Commenter

    I've noticed this issue also.

    0
  • Ian Wilson
    Moderator
    Top Commenter

    In my experience the way to get the linear gradient lines parallel in this situation is to make sure to drag out the gradient from the middle of the edge, not from one side or the other. 

    Ian

    1
  • SFA
    Top Commenter

    As far as my testing has shown, based on my understanding of your description, both gradient masks respond in the same way and are applied to the whole image before any distortion adjustments are applied.

    Reference points stay with the points of the image as lens correction is applied.

    What am I missing?

    0
  • Ian Wilson
    Moderator
    Top Commenter

    SFA - the problem is illustrated here. If you drag off centre, the gradient mask may be tapered if there is a lens correction. (You could argue that it is more sensible for the mask to follow the geometry of the uncorrected image, but you can see why users might not like that.) And it was meant to say "non-parallel gradient mask" but some words got lost when I resized it for the forum, I think.

    Ian

    0
  • SFA
    Top Commenter

    Hi Ian,

    Sorry, I had only seen the first post in the thread when I responded.

    I get the point but then that would require 2 different methods of applying the mask  - one following lens correction and one not - depending on the nature of the adjustment. Perhaps even a distorted linear mask, if such an idea makes sense.

    But my main point was that I did not really see any difference between linear and radial mask application in principle, as suggested by Lorenzo.

    0
  • Thomas Kyhn
    Top Commenter

    Here's a related request: Gradient masks after keystone adjustments.

    I asked support about the issue; here's part of the reply I got:

    We are aware that there is an issue with the Radial, Gradient masks, and the brush in relation to the image geometry. When using tools such as Keystone or Distortion (lens correction tool), these masks can become deformed as they are tied to the image geometry. We understand that this can be burdensome for our users.
     
    Please note that this issue has been present for some time (from almost the beginning) and changing it would require significant resources as it would involve rebuilding a substantial portion of how Capture One core functions in relation to tools and the image. Consequently, the cost of fixing this issue is quite high.
     
    However, we want to assure you that addressing this issue has always been our intention. We also believe that the impact of this issue is not very significant, as it is only noticeable when pushing Keystone a bit to its limits, which is not a common scenario. Additionally, there is a simple workaround for those who are affected by this issue, which involves applying the mask before applying Keystone.

    So basically, in the view of Capture One, the company, the issue is insignificant and would cost too much to fix.

    2
  • Lorenzo Ceva Valla

    Thank-you Ian but what I ma saying has nothing to do with what you say the problem is different!

    As I said the problem is that the linear gradient is distorted independently from how you draw it!

    0
  • Ian Wilson
    Moderator
    Top Commenter

    Can you show a screen shot to show how it is distorted?

    Ian

    0
  • Lorenzo Ceva Valla

    SFA

    In my experience there is a big difference in the behavior of the linear gradient and the radial gradient with files that have been strongly corrected for distortion with the lens profile.

    With linear you get the overlay graphic perfectly linear but the resulting gradient is distorted, with the radial gradient you get the opposite situation as the overly graphic is distorted but the effective gradient is correct.

    At list this is what I get with the 28mm of the Leica Q3 and with the 20-70 by Sony which are two lenses that need a strong distortion correction.

    Best regards

    Lorenzo

    0
  • Lorenzo Ceva Valla

    There it is!

    The first one is a linear gradient on an Leica Q3 with distortion correction activated as you can see the gradient is distorted.

    The second one same thing but with the distortion correction disabled and as you can see this time the gradient is linear but unfortunately I like to work with pictures with no distortion!

    Best regards 

    Lorenzo

    0
  • Ian Wilson
    Moderator
    Top Commenter

    OK, I see what you mean. It's curious, because

    • if you draw the mask in the middle, as I suggested, the lines are straight and parallel, but nevertheless the mask itself is curved, and
    • if you draw the mask away from the middle, you get both things - the lines aren't parallel (but they are straight), and the mask is curved too, but no longer symmetrically.

    I'd never noticed the mask being curved like that, though I suppose that in many cases it would make little difference: the mask would have a fairly gradual fall off and in any case a dead straight mask wouldn't always be any better. In my image, I used a gradient mask to lift the shadows in the foreground which are not straight anyway. I'd need a mask with a very gradual fall off so that it didn't look too obvious.

    Your mask is very hard edged, though it may be you have done it like that just to make the effect easy to see. 

    So I agree it's odd. Clearly the mask is following the geometry of the uncorrected image, even though the lines aren't. And it would make more sense for the mask to follow the geometry of the image as corrected. How much of a difference it makes in real use depends on how soft the fall off in the mask is, I presume. 

    From the reply that Thomas Kyhn had from Support, it sounds like there is a chance that they may do something about it eventually. (His case was about masking with the keystone correction active, but it's all part of the same issue, it seems.) Meanwhile, I suppose 

    1. in many images it won't matter, because you want a pretty soft mask anyway, such as in my image, but
    2. when it does matter you can try what was suggested to Thomas, and disable the correction while you draw the mask. The trouble with that, though, is that for correction of lens distortions, straight lines in the image may only be straight in the corrected image which is very obvious in the pews in the foreground of your image (do they call them pews in a synagogue?) so drawing the mask as required with the correction off may or may not be any easier.

    Ian

    0
  • SFA
    Top Commenter

    Doing something quite extreme I picked random but simple image with useful reference point and created series of variants.

    The original shot used  Canon 70-200 L F2.8 Mk 2 - a lens with relatively little need for any adjustment and therefore a good baseline.

    The alternative profile I applied to the shot using the Lens Correction tool was the C1 provided "generic pincushion distortion" with the distortion value set to 120%.

    For both linear and radial gradient masks the results were the same. 

    The mask definition was applied to the base pixels of the RAW file in an uncorrected state as far as distortion is concerned. The actual mask then mapped is presented using those pixels selected for the mask AFTER disortion correction.

    For lens profiling purposes that seems reasonable, especially for things like peripheral illumination correction and sharpness fall off.

    In most cases where adjustments will be quite subtle, the results should be fine.

    For extreme adjustment, especially if the accuracy of the mask is unrelated to natural lens-induced adjustment needs or exposure, colour or whatever, consistency adjustments, that might not work - although at the same time, it suggests that in fact, the basic Linear or Gradient mask might not, in any case, be the ideal tool for the correction required. A gradient mask with manual corrections might be closer to the requirement?

    That would suggest the parametric gradient mask would need to be rasterized and adjusted. Or perhaps a different approach would turn out to be more effective in some cases.

    As far as enhancements are concerned, maybe there needs to be some sort of Raster Mask creating versions of the Gradient masks?

     

    1
  • Lorenzo Ceva Valla

    Obviously my example is absolutely exaggerated but the distorted effect of the linear gradient tool is visible even with softer settings. 

    Obviously if the more the gradient is draw in the middle of the frame the less the distortion effect is visible as the distortion correction is stronger near the edges of the image.

    Using the gradient without applying the distortion correction is not an option as specially with some lenses the distortion correction is absolutely needed.

    Obviously this problem is visible especially with lenses that need strong distortion correction like the one on the Leica Q3 and also with the 20-70 by Sony and this effect is not a problem with lenses which are physically corrected.

    As I said I tried with Adobe ACR (which I don't use) and the linear gradient is perfect after the distortion correction as it should be!

    Lorenzo

     

    0
  • Thomas Kyhn
    Top Commenter

    I'm not sure about the exact consequences of the distortion correction issue described above, but keystone adjustments don't just affect the shape of gradients. One useful thing when placing gradients is the option to hold down the shift key to ensure that they're vertical/horizontal. This usually doesn't work when keystone adjustments have been applied, and it usually makes no sense to apply gradients before keystone adjustments have been applied, so this workaround isn't of much use.

    0
  • SFA
    Top Commenter

    Lorenzo,

    I see this as a different approach to lens correction by C1 compared to ACR.

    C1 have chosen to correct the lens problems and assume that whatever defects have been resolved need to be resolved at the base level (i.e. pre-adjustment) for all subsequent adjustments.

    ACR, based on your description, chooses to apply the adjustments to the post-correction output.

    In most normal cases of minor adjustment, this probably would not be very different., even to the point of not being visible and/or important.

    In the modern world one might expect the lens manufacturers to provide the required data correction, especially for massive distortion effects, along with the image but to rely on that might be foolish. Obviously, with older lenses and fully manual lenses, correction data is rarely, if ever, available. 

    Whichever approach one takes  - C1 or ACR - the area adjusted by the layer mask (or ACR equivalent) is adjustable either by adding or subtracting the masked area. if so desired. Indeed, such adjustments for gradient masks are not that uncommon in regular use since gradient masks per se are often not a precise solution to an adjustment need.

    Either way, to correct to one's choice the mask would need to be rasterized.

    It seems to me that the initial parameterized approach for the mask, when based on the inclusion of lens correction characteristics, is more likely to be generically useful for things like copy and paste to a selection of images. However, in terms of personal preferences, user by user, the choices may not be so clear cut. 

    0
  • Lorenzo Ceva Valla

    I am a very practical person. What i know till now is that in C1 with some lenses (on the q3 I can't change it -))) in some cases linear gradient tool is problematic while it's not on ACR. 

    To me it seams "normal" that a tool like the linear gradient should operate on the corrected image as it is the first thing you do and also because it's not a choice but I am obligated to do the distortion correction!

    I know very well that I can rasterize the mask and then adjust it with a brush and this is what I do already but we must admit that this is not very comfortable especially if you have to deal with hundreds of files!

    Will I start working my raw files with ADOBE ACR or LR? Certainly not! I work on C1 since many years and I don't want to change as I am very happy with the results!

    I understand that it might be difficult to fix this problem but I am sure that it will be fixed because C1 is a serious software and this behavior does not have any sense!

    0
  • Thomas Kyhn
    Top Commenter

    I understand that it might be difficult to fix this problem but I am sure that it will be fixed because C1 is a serious software and this behavior does not have any sense!

    Unfortunately that isn't always how things work. Capture One has had this issue for many years, the company knows about it, nothing has happened yet, and – as the reply from support cited above indicates – it's unlikely that this will be fixed any time soon. The same goes for many other old issues.

    0

Please sign in to leave a comment.