I’m looking for some expert opinions/advice regarding sharpening after resizing in Gimp.
I appreciate that there are almost countless ways to sharpen images but I quite like a strategy Pat David wrote about - decomposing the image to wavelets, copying the first and second scales and varying the opacity.
Here’s what I’ve done. My image started out at 5631x3754 and I’ve resized it to 690x460 in Gimp (cubic). I then decomposed to 6 scales using GMIC Split Details (Wavelets). I duplicated the first and second scales and set the opacity of the duplicated scales to 60% and 30% respectively.
And here’s what I’m asking. To experienced eyes, is this the right amount of sharpening for a resized image? Or, is it still too much?
The sharpened images both look bad, but even the original image looks bad because there is massive aliasing along the edges. The most likely cause is that the resizing algorithm used has a tendency to be over-zealous, so choose a different downscaling algorithm. Sometimes that also happens due to the jaggedness of the object - in this case the ropes might be plaited. Hard to tell without seeing the 100%-sized image.
If your intention is to increase edge acuity then USM or RL are fine for the task and simple to use. If the intention is to increase detail/texture then maybe something like EPD or wavelet local contrast (in RawTherapee see Wavelets > Edge Sharpeness and Final Touchup) are better.
I stopped worrying about detail when I finally pixel-peeped a 640x480 image. For my purpose, the very minimum USM does the needed job, to provide the illusion of sharp.
Yes, the rigging in the resized but not sharpened image looks bad, so any sharpening will make it look worse. You might post a link to the full size version.
To my eyes, the water and foliage look good in the final version, so you might use an adaptive sharpening that leaves the boat alone. But it’s better to sort out the resizing first.
I wonder why you used cubic. When you downscale, you can avoid aliasing problems by pre-blurring the picture by the same amount as your scaling (Gaussian 3px blur if your downscale 3x, for instance). This removes high-frequency components that could cause aliasing and wouldn’t be visible in the result anyway (since they would be smaller than a pixel).
Where are you seeing sharpening artefacts? In the ropes to the left of the front mast? Those do not look quite right to me, but I’m not sure if it’s from the sharpening or not.
Okay - great - now I’m getting it. I can explain the “lighter” bits. They were caused by me inadvertently double sharpening in Canon’s DPP software. I used the lens optimizer and Canon says to turn off the sharpening or it will cause artefacts - I hadn’t done this. So, I have now turned off the sharpening and the “lighter” bits seem to be gone to me.
@jules Definitely need to be wary of DPP because it could sharpen without you knowing about it. In camera, I normally try to make everything as unmodified as possible; e.g., setting picture profile to Neutral. You can do that in DPP as well. That way, you have more control when you use GIMP, etc.
So, with everyone’s input, I took a fresh TIF, scaled it down to 650x490 and applied Constrained Sharpening. In terms of downscaling and sharpening a downscaled image, is this now all right? (I know other things could and should be done to this image but for this exercise I wanted to focus on downscaling and sharpening.)
I cannot usually get the result I want just by applying simply one sharpening to the whole image.
Here is my common workflow regarding sharpening when dealing with photos publish in the web. I am shooting mostly portraits.
Raw-conversion in RawTherapee.
Apply some preferably RL-deconvolution “capture sharpening” if it is ok i.e. not too much noise or wrinkles (portraits). This way you can get a little bit sharper resized image.
I am mainly shallow dof / portrait shooter and that’s why local contrast is not applied at this point.
Resize to 2048px and export as 16 bit tiff.
Finishing in Gimp.
First consider local contrast (LC): if the pic asks for some larger scale pop, GMIC has some nice tools like Local Contrast Enhance and Texture Enhance. Choose the area in preview image which needs strongest LC settings, mess around with settings to get a good result and click ok. Remember that you can always pull down opacity if the effect is too strong. Usually you have to apply a layer mask and
a) paint black the areas where you don’t want LC (usually: bokeh, OoF areas, face, skin)
b) paint white the areas where the used LC setting is ok
c) use shades of grey in the areas where you want less LC
! If you use strong LC settings, be especially careful with black-white transition zone’s in masks. Use medium-soft brush. It is a good idea to use the free select tool to roughly set the area of effect, pour in a can of paint and finish the borders with brush.
! Sometimes you don’t need LC and dodge & burn is enough.
Sharpening is the final step before watermarking (if use one) and exporting. Nowadays I prefer GMIC’s RL-deconvolution for sharpening and it is done the same way LC is applied: pick up the area which needs most sharpening, get the settings, apply layer mask and paint:
a) black: skin, bokeh, OoF, etc. you don’t want sharpened
b) white: eyes, eyelashes, clothes, props, and so on
c) grey: lips, hair, …
! Sometimes you have to create two sharpening layers and use masks to get separate them
=> Export and be happy <3
In your case I would use mask and ease sharpening on ropes and other areas where you get jagged lines with full setting.
And yes, I am aware that some things like applying LC to bokeh is kind of matter of preference / taste, but I am here just sharing my opinions. Feel free to disagree.
As an example, below is one portrait edited this way and a screenshot from Gimp showing layers and masks. Didn’t get rid of all jagged lines in her sunglasses, but otherwise sharpening is ok to my taste.
