I shoot on a D300 (~4200 x 2800 pixels). I open my raw images full size (via ACR) and then do whatever I’m going to do on the full size image in Photoshop – layers, correction layers, filters, etc.
Once I’m done I save off a copy for my web-based slideshow/gallery system. Those images are 625 x 480 max.
– If I downsize images in a single step (using Image Size or via Save for Web [in either case set to Bicubic Sharper]) the result looks a little soft.
– If I do it in two steps (first step to about 1/3 size of the original, second step the rest of the way) the result looks a little crisper.
– If I do it in three steps (reducing by half each step) it’s even crisper, but starts to look a little chunky/jaggy, or perhaps oversharpened.
I know the whole "resizing in steps" thing was supposedly rendered obsolete by Bicubic Sharper, but it seems to me that it still delivers better results than doing it in a single step.
My question is – has anyone done any kind of definitive (or at least somewhat more objective) study of this and come to any conclusions? I would love to know what the most optimal way of doing this is. Besides just "whatever seems to work best for me".
Learn how to optimize Photoshop for maximum speed, troubleshoot common issues, and keep your projects organized so that you can work faster than ever before!
Nope, it was made obsolete by Bicubic Smoother, for upsampling.
Oops – well, I was close – sorta…
What you could try is opening a second copy of the file through ACR and choosing a smaller size in the ACR workflow options.
That would be a good solution if I wasn’t doing any post-ACR processing. But since these images can have multiple end destinations – Epson print, small web image, large image for stock – I want to do all my work at a relatively high resolution and then save off copies for various destinations, rather than redoing it multiple times. Not really an issue with global adjustment layers, but once I get into masking or spot retouching it becomes one.
Then I would definitely go with one-step downsampling in Bicubic Sharper and apply appropriate output sharpening at the end. I use PhotoKit Sharpener by Pixelgenius myself for all my sharpening.
PhotoKit Sharpener looks pretty cool. Does the sharpening as separate layers (or layer groups) – right? So you can disable/enable different applications.
Still wonder though… if the degree of resize (i.e. how significant the size change) affects how much the image is softened (which seems to be the case), if you do a single step resize, don’t you have more de-sharpening to remedy in a single step? Vs. if you do multiple smaller steps, the amount of softening per step is less, meaning less aggressive sharpening is required in each step to pull the image back to sharp. Which makes me want to ask – how does Bicubic Sharper actually work? Does it resize and then sharpen? Sharpen and then resize? Or is the sharpening a built-in by-product of how Photoshop analyzes groups of pixels for resampling? If one of the first two, seems like you could manually do the same thing by resizing and then sharpening, or vice versa – and with greater control.
I suppose this is all academic – I don’t have the tools or skills to be able to definitively compare the various ways of getting from A to B.
Could be an interesting exploration though, for someone so inclined.
If you feel like bicubic sharper produces soft results, then you might try some unsharp mask when you’re done sizing down (one step down, not 2, 3, 4). You could also try instead duplicating the background, running filter… other… highpass at 0.3 to 1 pixels and setting the highpass layer to overlay or soft light. Reduce opacity to suit.
Sharpening (with unsharp mask) is also best done on a duplicated layer. You can then set the layer to darken to avoid the light halos that create that "over-sharpened" look. Maybe dupe that layer and set to lighten with less opacity.
Take care that your display is not the softest or the sharpest on the block. Your best bet will certainly be to preview your images on as many displays as possible to be certain you aren’t over- or under- sharpening.
PhotoKit Sharpener looks pretty cool. Does the sharpening as separate layers (or layer groups) – right? So you can disable/enable different applications.
Yup.
"Multiple smaller steps" of resampling is just a bad idea. You’re multiplying the artifacts with each step.
Yes, Ramon. I find for web sharpening though, controlling the light side of the halo (which is also easily done with luminosity blend mode and blend if) is an easy way to process an image sight-unseen. I have read and would certainly recommend Bruce Fraser’s Real World Image Sharpening.
Cool – tried both methods – highpass and USM on another layer. Both work great. Highpass is a little less intuitive – I’m more likely to stick with USM just because I’ll remember how to do it 🙂
Seems that if I’m judicious in my USM I don’t need to set the blend mode to darken. The blend mode change really reduces the amount of apparent sharpening, so I have to hit it extra hard, then back it off via setting blend mode to darken. Or I can give it a light tap with USM and not change the blend mode. Since I’m not asking for a lot from the sharpening – just to tighten up the image after a big size jump – this approach seems to work.
Luminosity mode looks the same as normal, at least to my eyes. But I see how I can pull back the white on the upper Blend If slider to reduce the white side of the halos.
Anyway, this seems like a much better way to control the exact sharpening than doing a multi-step reduction.
If you convert the layer to a smart object, you can always adjust and readjust the sharpening as needed, as well as change the blending options and their opacity right in the sharpening filter by right clicking on it. It’s non-destructive and negates the need to have duplicate layers.
The whole thing about blending sharpening to Luminosity was created originally by Dan Margulis. Sometimes has its uses, but its not necessary as a default, nuff said.
Actually, blending or fading sharpening to Luminosity arose as an alternative to converting to L*a*b just to do the sharpening in L*, which was what folks like Margulis and Kelby were advocating at one time. (Maybe they still are, I don’t know.)
