The Sharpening Thread!

[livin4lax09]

Here we go, PJ suggested it, so it has arrived! I'd like to see what others are doing, and it would be helpful to share your techniques. So here goes. Post whatever sharpening techniques you may use. Here goes mine...

No Unsharp Mask resized to 800 pix wide



Unsharp Mask of 500% radius .3 threshold 0, then resized to 800 pix wide



Unsharp Mask of 500%, .3 radius, 0 threshold, then resized to 800 pix wide, then 100%, .3 radius, 0 threshold






and here at 100% crop

No Unsharp Mask



Unsharp Mask of 500% radius .3 threshold 0

[OmahaAdam]

Concerning the 2nd and 3rd photos...What's the logic in usm once, resizing, then usm again, a la the 3rd photo? I can see there are differences between the two methods, in the texture of the blue jersey and goosebumps on the leg for example, but why not just usm once after the resize? Not criticizing, just curious.

[drg]

Brent,

I believe you will find that the sharpening technique you are using looks better perhaps than images that aren't sharpened at all on first glance, but...

The point of the USM tool is to sharpen selectively the image. Using a threshold of 0 in Adobe Photoshop sharpens all the pixels. You can see in the jersey lettering the artifacts that get introduced by this technique. Skin gets badly treated under many lighting conditions by not having some threshold, and even noise reduction won't always fix the color flaws that can occur.

I like to de-blur/de-haze the image when using USM . I first use a amount of around 16-20%, a radius of 40 (yes, 40), and a threshold of 1-3.

Then, I go back and use the Unsharp Mask, and more conventionally sharpen:

Amount 100-150% range (without people and faces or text I will go higer on occaision)
Radius .7-1.1 (1.5 or higher starts to create over big halos in high contrast areas)
Threshold 1-3 (with people) and up to 5 or 6 without people.

I will dig up some examples later this pm.

There are a couple of other sharpening techniques that include tracing edges and a newer one, I'm not really facile with yet, which involves playing with the registration.

The registration is a new technique for me (the print folks and some GD's swear by it for big enlargements). This results in crisper photos than fractals will produce alone.

One other quick note:
As I've done more and more work in the 'digital' and 'web' age, there is a big difference between printed and electronically displayed images as to what works. Light Transmitted vs Light Reflected.

Interesting thread and I look forward to also hearing (I'll not comment on this topic yet) what others consider too much sharpening.

[masdog]

Brent,

You have an interesting technique, but the images almost seem TOO noisy. Maybe its just texture, but thats how it seems to me.

I don't know my USM settings off hand, but they're similar to the second pass that drg posted. His dehazing step seems very interesting, and I will have to give that a shot sometime.

[livin4lax09]

sean, the 1d's shots tend to look like that, as the detail isn't as smooth, since after all, it still does have 4 MP, no matter how good the pixels are, there's still only 4 million of them.

DRG, as PJ put it, there is not one specific sharpening technique as an answer-all. I tried your technique because I was curious, but it left me with less than desirable results. The different type of photography also comes into play. and of course, preference among all else. I have played around with many different USM settings in the past couple years, and this has been the one I arrived at, for the 1d. my settings for the 300d were different as well. your first step, the dehazing, would work much better for a shot taken in controlled lighting, but it tends to boost the contrast, and at that point, i start to lose my brights, especially under harsh sun like this.

Omaha, I'm not exactly sure the reasoning behind this, but in my experience, I have found that you can get more details doing it this way, in the two-step. I was told about a year ago on another board that my photos were too soft, and that after I resize them I should sharpen again, so I did and it has improved them a lot. And another reason I do it in two steps is because of my workthrough. I edit my shots for what i want the print to look like, and in that step, get everything down correct, including sharpening. Then when i decide i want to put them on the web, I have to resize them and at that point sharpen again.

oh and also I should point out that it matters where you get your images printed as well. some printers tend to print soft, while others harsh. find the medium ground for your printer. This is why i use this technique, because my lab prints these with just the right amount of sharpness at these settings.

[drg]

It is a good point that there is no one perfect way to sharpen. There is though, a need for a basic starting point for approaching this the first few hundred (thousand times) when using Unsharpening (USM).

The de-haze method I described itself has lots of variations, particularly in the threshold region and depending upon the image itself, but the reason for its success is that it is an additional noise reducing technique because of the interaction of the high radius and lowered intensity. It works as a Very Smart Blur (my caps). Some images it is unecessary, or adds little. It is an addition to a basic starting point for USM and isn't always a good starting technique.

Yes, these methods alter the contrast (look at before and after sharpening histograms) but that is exactly one element of how USM works. It changes the visual perception by altering the accutance of the photograph by inserting highlights, also known as halos, in place of other pixels which definitely changes the overall contrast of the photo! This is not a perfect description, I'll work on better one if you want:idea: (had to try that smiley)

The two step process is as much about the effect of resizing as it is about the USM method. When you re-size, you alter the original relationship of the visual data, and sharpening addresses that via altering the transition between pixels so they look better based on one of many algorithms (there's a $5.00 word).

Simply put, when you re-size, there is new information (or less) that has to be fixed to look as good as the original.

My separate post in this thread, The Puppy, displays a worst case scenario, a black dog against a horrid background in bright sunlight. That post speaks for itself with what I did and didn't do to 'fix' the photo and displays some of the techniques so far discussed.

Back to the what I am going to call the %500USM method: this is essentially just heavy sharpening, and may be working for some pictures and printing methods (though my graphic designers can't figure out how anybody could have so badly screwed up any variety of printer/processor/inkjet, etc. to make this work) but more often than not it is going to introduce unacceptable levels of noise.

I am still interested (and we have examples in this post) of what others have to say about oversharpening.

[drg]

[There is a previous post to this if you are just looking at the most recent posts that relates to this set of images]

The photo used is a crop of a resized image.This is to illustrate based on where sharpening and adjustment are often used. The first composite shows the image I started from, centered, which is a closeup/crop of a larger image, and various effects on crops around the perimeter of the main image.

The second puppy pic, shows the final effect of Jonathan Sachs implementation of an Advance Sharpening - Noise Reduction method. I feel this variation came out the best.

The HS and 500 samples (Heavy Sharpening is also known as Sharpen More in some PS or other editing programs variants) show little variation, certainly at this scale. The increase in noise is very apparent.

[drg]

Fuji Film - Astia 100 slide/chrome original. Gallery full sized original.



Sharpening is not an exact process.

When sharpening, the goal is to improve the photograph, and not degrade it. Oversharpening is one problem area, but so is adding noise. By careful (or experimental) application of two techniques, dramatic improves in an image can be made without serious degradation of the photo.

Finding Edges in a photo (edge,edge trace, find edge, glowing edge, etc. are some variant
filter names) is one element. The other is selective sharpening. You can select a portion of the photo via masking, Selection (lassos and Magic Wands), but you can also select only certain components of the image!

The adjustment of only the the Value channel in the HSV model, will prevent most color noise and degradation. It will appear different (vastly) in some cases, but noise is not introduced into the color. Where Color Noise is already a problem, this method combined with a NR tool may be the key to producing the photo you thought you made in the first place.

One thing that basic USM in Photoshop does not do without help, is to accentuate just the
edges of elements in images. The edges are where the most effective sharpening
occurs in a photograph. So to help along this process there are some techniques and
third party tools/filters/applications. This is a sampling/overview of results that can be obtained with a photograph that are not just stock built-in variants.

These sharpening methods are based on finding edges of the Value Channel in the HSV model. For this image, already color compressed by jpeg conversion, using the Lab color space would have been overkill and on testing didn't make a lot of difference. Thus a very small portion of the image is actually touched by what we think of as sharpening. Just the edges of one channel (about 30-40% of the data) get sharpened by these techniques.

In these examples the two are combined in varying ways with traditional sharpening methods to accentuate a simple flower picture that was originally posted merely as a sample scan to evaluate film grain.

All editing of this type works best on tiff or non-compressed originals from RAW. The jpeg file has it own problems and artifacts, but this flower, not best case scenario, is a good example of what might be achieved with an otherwise throwaway file. This flower image, was part of a catalog shoot and this scan is a very small sample of the original. It has absolutely no adjustments made to it beyond what happens in the scanning process to achieve the best D-Max profile.


All work was done on the linked gallery file!

The original gallery file was adjusted as follows:

1. Speck Removal (custom healing brush), it is a film original afterall!

2. Removed my Signature (for resizing) by cloning with an oval soft-edged brush.
Smudged area to 'clean up' few remaining traces of cloning.

3. Resized image to 339x400 pixels for convenience.

4. Separated the image with Channels Palette in to HSV components.

[ There are several steps here regarding use of Layers, Channels, and Masking that are barely touched on and thus this is not an exact set of formulas. I am intending to demonstrate the results of sharpening variations. ]

5. Found edges 'Filter>Stylize>Find Edges' at 21% of the Value and lightened
the effect in a Brightness layer.

6. Created a mask (flower portion to mask background/foreground in samples 2-5)

Each of the following adjustments as described resulted in a sample file numbered below.

The adjustments via unsharp mask filter (USM) were done on the active layer representing the Value Channel (VChan) of the original image and then the image was saved. That image was then cropped with the cropping tool with margins of 151 pixels from the left, 5 p. from the right,5 p. from the top, and 179 p. from the bottom for a final cropped 'sample' image of 183x216. This was set up as an action as it was performed repeatedly. The whole image was additionally retained.

1. The original resized image for comparison.

2a. VChan masked and background darkened completely to lift the flower.
USM 50% r-1 t-3 applied to 'active non-darkened region'.
Mask made by filtering the the channel through high contrast filter to get
the black and white mask that was used.

2b. Same VChan as above but brightened the background 25% and
darkened foreground -40%. Just reversed the application of the mask
in a Duplicate Layer. Gaussian blur Radius of 9. Some Amount tweaking to acheive
a 'look'. This is really not an exact process. The background has to look good and
the bokeh should not be drastically changed. The layer percentage was set to 100% for
the overall effect.

2c. Mixed the two channels backed together to produce the new VChan. Then the
original Hue and Saturation channels were combined with the new Value channel to
produce the image.


3. Image Sample 3 has more of the same amount of USM from 2a. to demonstrate
a degree of over sharpening. The USM was applied to VChan before mixing
the channels back together again.

[There are some edge artifacts in the two example from 2 & 3 that are from
not making the mask exactly, and they appear as a redder/magenta-ish
color along the edge of some petals. The mask was quickly done with HIGH contrast
filtering and some minor brush work]

4. 10% reduction in shadow area brightness and 50% edge sharpening via channel
masking of VChan.

5. Noise reduction filters (I think it is Noise Ninja that has been plugged into a
pull down menu as a 'default' via a development kit, I'll look and make a note it later)
and Nik effects with a Sobel derived algorithm for sharpening have been applied via
masking on the VChan.

6. USM %56 r-6 t-16 to VChan

7. USM repeated on VChan - same settings

8. USM 100% R-1.0 T-2.
9. USM 17% R-40 T-3(The de-haze/NR USM application) and second USM 100% r-1.0 t-2.

Samples 8 and 9 were adjusted by applying USM to the original crop, and not via the Channel method.

[drg]

The final results non-cropped of selected images from The Flower - Part A.

First is the starting image #1.

Second is # 2 from Part A - the multi-masking technique

Third is #5 from Part A - 3rd party noise reduction and sharpening.

[Copy_Kot]

This is a great thread!!! I hope it doesn't fade away. DRG, thank you for doing all of that work (this newbie appreciates it).

[readingr]

Hi all,

This is fascinating and I have written a resize for the web based on something I found on the web.

In this I resize photos by several steps with USM. The resize is done in 25% steps till it gets to the right size. Always had problems up until I found this technique with artifacts being introduced. The USM I use is 150%, threshold 0 radius 1.

The script I wrote is for PSP and includes things like adding signature, flattening and change to sRGB from Adobe.

Must try the advanced technique drg pointed to.

Roger

[Medley]

Sharpening is one of those things that should always be done on an image by image basis. Here are a few points to keep in mind:

Sharpening can cause color shifts. A good way to minimize this is to duplicate the layer, sharpen the duplicated layer, and set the blend mode of the layer to Color. This utilizes the detail from the sharpened layer, and the color from the underlying layer.

Evaluation of sharpening should initially be done at the 100% or "actual pixels" view to prevent under or oversharpening.

Running the High Pass (Filter>Other>High Pass) filter on a duplicated layer at a radius of 5-10 pixels and setting the blend mode of the layer to Soft Light is an effective way to sharpen some images.

If you have excessive haloing on either the light or dark side of the sharpening, try this: duplicate the background layer a second time and re-apply the sharpening with the same settings. Set the blend mode of one layer to Lighten and the other to Darken. You can now adjust the light and dark halos separately using the opacity sliders.

A good overall formula for shapening is this :set the threshold to the desired level.( entire books have nee written about this one setting, so we won't get into it here. Next, set the amount to 500%. Adjust the radius up to the point where the image begins to break down, then reduce the amount until the image looks good.

-Medley.

[Sushigaijin]

this is a great thread, glad I found it...

I'm going to hesitantly agree that pre-resize sharpening has it's merits...recently I've been using high pass sharpening at a very small radius, usually 3-4 px on the full size image before resizing - the amount of highlight detail depends on if i use overlay, soft light, or hard light. The small high pass radius is intended to sharpen the hard edges but ignore the details - most of the time I think that details (like catch lights) look oversharpened even before sharpening, no need to sharpen them twice. The USM used on the resized image is usually a MUCH lower amount than I would have needed had I not sharpened already...always good to need less sharpening.

here's a shot that I found in the garbage bin, the detail is great but there are a lot of bright highlights to worry about, notice the dust and the hair...all compression is equal in these images.

just resized:

Just high pass sharpened, then resized:

Just USM, no high pass:

high pass, resize, USM


I'm not sure if it is that obvious on this example, but on soft pictures it can really make a pretty big difference.

I am still looking for a good workflow for sharpening full-resolution images, everything looks pretty good at 700 pixels but rarely am I totally happy with the sharpening of the exact same image at 7 million pixels. Any advice?

Erik.