Praise, wishes and the n-word
First some praise:
Congratulations for the keystone correction tool. It seems to have problems with extreme keystoning (worm's-eye view with a 14mm) but within reasonable margins it works wonderfully. A real treat!
Also the local corrections are most welcome. Just add a gradient tool, please. I know I can achieve similar effects with large soft brush settings but a gradient is simpler and faster in use.
Being able to adjust the size of the local adjustment brush and the spot removal tool with a right mouse click is very nice but when doing so the scale for the brush size hides the image at the very point where I want to use the tool. LR's approach with the scroll wheel is simpler and faster in use.
This leads me to another ease of use I would like to have:
Typically I want images with no distortion, no chromatic aberration and no vignetting. Only for the very few occasions when I deliberately would like some of these lens imperfections to show up in my image I would apply manual adjustments. Right now with C1 it's the other way around, at least if you are working with DSLRs (which I do). LR, Capture NX2 and even the in-camera-jpgs automatically recognize lens, focal length and aperture and correct the above mentioned lens faults (of course with the option not to have them corrected fully or partially). I'd like this in C1 as well because if you are shooting outside a studio or similar situation, lens values change with almost every shot making the manual synchronisation of settings very time consuming. It would not even be necessary to profile a lot of lenses. The data of PTLens are already there (not perfect maybe, but good enough for most cases as a start). But practically it would be sufficient to have correction profiles for Canon's L lenses and the Nikon pro glass to start with - ok, maybe for my S95 it would be nice to have it as well.
The notorious n-word: noise
In my perception the main reason for the issue of "dirty" noise is the sharpening algorithm. Without sharpening the image remains smooth and amorphous. Once sharpening is applied, very dark or very bright pixels appear, often as a string with a length of up to six or seven pixels, frequently two strings connected to an angle and always exactly horizontal or vertical. They are clearly identifiable as artifacts even to the untrained eye and contaminate otherwise clean areas because of their regular shape, position and hard contrast. Apparently this effect happens to be even more articulate with Nikon's NEFs than with Canon's CR2 files. In LR the sharpening artifacts are amorphous and have a lot less contrast in areas with no texture. It appears to me that C1's sharpening algorithm needs to be worked over because if I take an unsharpened C1 image to LR and sharpen it there, these artifacts do not appear.
The second issue is color noise or high ISO color noise to be precise. Up to a color noise reduction value of about 20 to 30 everything is pretty much under control. Above 30 colors start to blur more strongly causing a visible desaturation of small areas and color-only contours start to disappear. But at these values of color noise reduction the blue blotches often have not disappeared (and color noise is mostly about these blue speckles especially since most high ISO images are taken at tungsten lighting). At values above 40 the luminance structure in she shadows starts to disappear resulting in very blurry areas. The automatic values for color noise reduction are often ridiculously high causing areas with low contrast to become untextured even in the luminance channel. One could disguise this by adding â€fine grain" but now things lose their simplicity which is not C1's way of doing it as far as I understand. I want to reduce noise instead of adding even more. So my wishes for color noise reduction are: an algorithm that recognizes high ISO noise and therefore does not need to blur the contours of colored areas and a way of correcting it that does not affect luminance texture when only color noise is to be addressed.
Other opinions, similiar evaluations?
P.S.: I'm planning to post this as well at the luminous landscape forum with samples that support my observations.
Congratulations for the keystone correction tool. It seems to have problems with extreme keystoning (worm's-eye view with a 14mm) but within reasonable margins it works wonderfully. A real treat!
Also the local corrections are most welcome. Just add a gradient tool, please. I know I can achieve similar effects with large soft brush settings but a gradient is simpler and faster in use.
Being able to adjust the size of the local adjustment brush and the spot removal tool with a right mouse click is very nice but when doing so the scale for the brush size hides the image at the very point where I want to use the tool. LR's approach with the scroll wheel is simpler and faster in use.
This leads me to another ease of use I would like to have:
Typically I want images with no distortion, no chromatic aberration and no vignetting. Only for the very few occasions when I deliberately would like some of these lens imperfections to show up in my image I would apply manual adjustments. Right now with C1 it's the other way around, at least if you are working with DSLRs (which I do). LR, Capture NX2 and even the in-camera-jpgs automatically recognize lens, focal length and aperture and correct the above mentioned lens faults (of course with the option not to have them corrected fully or partially). I'd like this in C1 as well because if you are shooting outside a studio or similar situation, lens values change with almost every shot making the manual synchronisation of settings very time consuming. It would not even be necessary to profile a lot of lenses. The data of PTLens are already there (not perfect maybe, but good enough for most cases as a start). But practically it would be sufficient to have correction profiles for Canon's L lenses and the Nikon pro glass to start with - ok, maybe for my S95 it would be nice to have it as well.
The notorious n-word: noise
In my perception the main reason for the issue of "dirty" noise is the sharpening algorithm. Without sharpening the image remains smooth and amorphous. Once sharpening is applied, very dark or very bright pixels appear, often as a string with a length of up to six or seven pixels, frequently two strings connected to an angle and always exactly horizontal or vertical. They are clearly identifiable as artifacts even to the untrained eye and contaminate otherwise clean areas because of their regular shape, position and hard contrast. Apparently this effect happens to be even more articulate with Nikon's NEFs than with Canon's CR2 files. In LR the sharpening artifacts are amorphous and have a lot less contrast in areas with no texture. It appears to me that C1's sharpening algorithm needs to be worked over because if I take an unsharpened C1 image to LR and sharpen it there, these artifacts do not appear.
The second issue is color noise or high ISO color noise to be precise. Up to a color noise reduction value of about 20 to 30 everything is pretty much under control. Above 30 colors start to blur more strongly causing a visible desaturation of small areas and color-only contours start to disappear. But at these values of color noise reduction the blue blotches often have not disappeared (and color noise is mostly about these blue speckles especially since most high ISO images are taken at tungsten lighting). At values above 40 the luminance structure in she shadows starts to disappear resulting in very blurry areas. The automatic values for color noise reduction are often ridiculously high causing areas with low contrast to become untextured even in the luminance channel. One could disguise this by adding â€fine grain" but now things lose their simplicity which is not C1's way of doing it as far as I understand. I want to reduce noise instead of adding even more. So my wishes for color noise reduction are: an algorithm that recognizes high ISO noise and therefore does not need to blur the contours of colored areas and a way of correcting it that does not affect luminance texture when only color noise is to be addressed.
Other opinions, similiar evaluations?
P.S.: I'm planning to post this as well at the luminous landscape forum with samples that support my observations.
0
-
Version 6.1 clearly addresses the mentioned color noise issue. As far as I can see the blur of the red channel kicks in smoother and at much higher settings of the color noise reduction slider than before. That makes this tool much more usable and leads to a lot more usable images, especially at low color temperatures. It does not dethrone LR3 from being the high ISO champion though but it's a clever move for a substantial improvement of this tool – as long as it does not recognize and eliminate high ISO color specs like the spot removal tool removes and eliminates spots. Thank you, developer team!
This gives hope for a similar kind of cure for the rectangular sharpening artifacts.
By the way, is there an option to display ISO, f-stop and focal length (or maybe even lens) in the browser thumbnails? They only show up in the list view which takes a lot of screen space (since the column widths are not adjustable) and the small images are too small even at the largest zoom setting.0 -
My perception is that the LNR slider changes it's degree of impact with the ISO value. Unfortunately I didn't find any written hints on that. At high ISOs heavy smearing of luminance structure occurs. So just in order just to try something I set it to 0 and tried the "surface" slider in the advanced noise reduction tab alone. Voilà , there was the effect I was looking for in the first place. The manual advises us to use this silder "To smoothen image surface without loosing sharpness". Well, that is exactly what I expect the luminance noise reduction tool to do. And I don't understand the veiling name for it. The classic LNR slider would probably need a new name too then. By the way: The video tutorial on this tool on youtube and the one from Walter Borchenko in Michael Reichman's C1 video tutorial are at least misleading regarding the power of this tool for high ISO noise. Didn't anyone try that out before?
Try it yourself, set LNR to 0, crank the "surface" slider to 25, 50 or 75 and see yourself how well this tool works creating a nice amorphous structure for high ISO luminance noise. It even does a pretty good job weakening the linear rectangular sharpening artifacts I was talking about in my first post in this thread. I know, these artifacts are not that pronounced in print as they are on screen. But my customers usually view my images on their screens, so this effect is not as unimportant as some users say.
To me these artifacts look like if there is an algorithm at work that takes the differences in vertical and horizontal direction into account but leaves out the diagonals, maybe in order to keep the calculating effort lower. In the early days of digital photography processing power was very limited compared to now. Perhaps this is some ancient code at work. We'll probably never know…
Now I would like to set my "surface" and LNR values as default but the "save as default" option is not really documented. Is it default for the camera model or the ISO? Having good defaults is a key to a fast workflow. That's why I would like it also for the lens correction tool when C1 is used with DSLRs.
One more aspect of the NR tools: Apparently NR (and sharpening amount, I'm not sure for the radius) is performed with the very pixels displayed on the monitor instead of image pixels. This results in not just unexact but completely misleading (blurred) views when zoomed out. With the calculating power of today's computers and graphics cards this is no longer necessary I think. Yes, there is the focus window but seeing the true (scaled) effect in the main window would make things less complicated.0 -
Apparently the default values for high ISO noise reduction have been altered in version 6.1.1. The classic luminance noise reduction is limited to 25 therefore the surface noise slider moves up to 20 and the "artificial noise" slider is also involved automatically. Still not quite to my personal taste, the rectangular "dirty" artifacts in untextured areas remain a yet unsolved issue to me, but again a clever and necessary move in the right direction. It would have been worth being mentioned in the release notes. 0 -
C1 6.3.3 improves the display of noise reduction on the screen when zoomed out. Apparently the NR settings are no longer simply performed on the displayed pixels but on the real pixels and there is an algorithm to give a pretty good impression of the real pixel NR in the zoomed out view. Thanks developers! Another weakness of C1 is eliminated. 0
Post ist für Kommentare geschlossen.
Kommentare
4 Kommentare