You may find that choosing steps in file size involving prime numbers - or at least ones not divisible by 2 - to give you less halo and artifact issues.
Jim
Bifurcator wrote:
But this TIFF image has already been sharpened. So more sharpening (or trying to show a different method all together) just messes it up. Without posting the RAW this is senseless IMO.
Capture sharpening is standard. It will be negated anyway when you downsize.
Btw, your idea of "microcontrast boosting" isn't bumping clarity to 100, I hope?
@Jim: You probably mean image size and not file size?
Resizing with Nearest Neighbor, the simplest of all.
First, interpolate sligtly down or up to a pixel number that can be divided by the final pixel number. In this case, 4 x 1280 = 5120
Then add gaussian blur with a radius corresponding to 0.4 pixels in the final size. In this case, 1.6 pixels. This is the Anti-aliasing filter, and the cutoff frequency can be chosen to taste and subject matter. From 0.3 to 0.5 px in final size looks ok.
Then resize to 1280 with NN.
Now we can remove exactly the blur we added, with smart sharpen -> remove gaussian blur, 0.4 px, 100%.
Finally, output sharpen with USM radius 0.2 pixels and amount after taste. Here I used 100%.
thanks for the link and examples! the difference between our results is pretty minimal in my eyes, i see a minimal difference in small structures like the window bars, mine are a bit "bigger". Wouldn't have thought that two so unrelated approaches would yield such similar results.
@bif if we based it on the raw we would compare raw converters and personal taste, not re-sizing algorithms. Anyway, here is the unsharpened tif, i am courious about your method
@alundeb: Interesting, but for my taste that's too sharp and crunchy. It reminds me a bit of the 100% view of an M9 file. IMO the edge contrast is too high everywhere.
Philip: Regarding similarity of different approaches, you can't fool signal theory and the resulting detail content should be similar if the method is good. But I must say that I am pleasantly surprised by the blur-NN method. In fact I think it looks as good as Lanczos.
Sharpening is largely a matter of preference, but downsizing wile retaing highest possible fine detail without aliasing is the real pain in my experience.
AhamB: That is a very interesting feedback, that the edge contrast is too high to your taste. I can in fact simplify and remove the last step with USM. In any case, it is interesting to have as reference.
carstenw wrote:
Sigh. I look forward to owning a 27" or 30" Retina display. Then all this stair-stepping stuff will be much less important.
on the downside you would have to show really printable files then. Not that i worry about someone stealing my pictures, i don't think that one could make any real money with them, but i would have a bad feeling when uploading a 6mp file.
Oversharpening at FM is as bad as the HDR-frenzy was a couple of years ago. Take a look in the Zeiss or Leica threads, not all but too many. In particular landscape photography dont need this. I still do it myself, even some HDRs.
In Phillips beautiful capture, take a look at artifacts in diagonal mast wires, poles, windows, or when a leg, arm or shoe consists of 2-3 bright white pixels on a dark background. Doesnt matter which version you prefer. Beauty feels a bit gone.
To be honest, I can't be bothered to resize/sharpen the images I post on the web; they're usually full-size. I know they're not optimal for presentation, but I essentially treat the full-size jpgs as an extra level of backup of images that I care about, so that's more worth it to me. And it saves me from having to run every image I post through Photoshop (if there was a way to trigger a PS script directly from LR during export, I might consider it )
Sorry if that's off-topic; I have enjoyed and learned a lot from this thread.
wfrank wrote:
Oversharpening at FM is as bad as the HDR-frenzy was a couple of years ago. Take a look in the Zeiss or Leica threads, not all but too many. In particular landscape photography dont need this. I still do it myself, even some HDRs.
In Phillips beautiful capture, take a look at artifacts in diagonal mast wires, poles, windows, or when a leg, arm or shoe consists of 2-3 bright white pixels on a dark background. Doesnt matter which version you prefer. Beauty feels a bit gone.
I understand what you mean. But it is not really artifacts in the image. They are correct representation of the reality within the constraints of few pixels. It is the image that is so free from blur that you can see the pixels on your monitor. It is your monitor that is to blame! (Carsten's point) Don't get yourself a foveon sensor
AhamB wrote:
Capture sharpening is standard. It will be negated anyway when you downsize.
Btw, your idea of "microcontrast boosting" isn't bumping clarity to 100, I hope?
@Jim: You probably mean image size and not file size?
Yes. I view pixel dimensions as file size...just area rather than megabytes. They are both size in the end, right? I don't need to step on a scale to know that the size of my waist should be smaller.
Here's my attempt at sharpening this
from the unsharpened tif, resized to 2134px on longer edge, bicubic, frequency based sharpening applied, then resized to 1280px on longer edge followed by my sharpening routine again.
Shrug, I never do anything special when I am "finished" editing.
I just have Bridge send it to PS via the Image Processor, change color profile to sRGB, jpg file type, and re-size.
The surprise for me is that the difference between my way of doing it 5 years ago and how I do it today is close to zero. I think I have to pay more attention about how I do this, maybe follwoing some of the recipes you have shared here. I would like to get more details visible at some places of the image but still not getting halos at all.
The method I use today includes downsizing in steps (50% at a time) with some slight Smart Sharpening (100% 0.1) until the last resize where downsize to the desired output size. There I use typically 40% and 0.4 in Smart Sharpening and finally I remove some small halos (having saved the original layer through the process).
For the record, I stand by my GaussBlur - NN - De-GaussBlur method without final USM, that will be my post #8 above. In fact, it will go into my workflow.
alundeb wrote:
For the record, I stand by my GaussBlur - NN - De-GaussBlur method without final USM, that will be my post #8 above. In fact, it will go into my workflow.
Correction; the crops aren't 100% crops of course, just crops from the 1280 pixel wide images.
I'll have another look #8 above. I like your results and will see if I can make the same with a couple of images.
