Oversharpened? Really?

[FrankMi]

I've seen the oversharpening issue mentioned enough times that I am starting to wonder exactly what it means. Is oversharpened a personal taste or a definable point which anyone could identify by specific characteristics in the image?

When two people look at an image and one says that it is oversharpened and the other says that it is not, is there a way to define the truth? Or is it so subjective that we just shrug our shoulders and agree with one or the other?

[jiro]

I'd say it's subjective, Frank. In some cases, when your eyes has a hard time looking at the image due to very high contrast rendition (similar to "oversharpening") then I'd say it's way oversharpened. There are certain images that will benefit with this style or effect but still moderation is advised to make it look acceptable to the masses. As what they say "you cannot please everyone the same."

[revi]

I think it's partly a matter of taste, but over sharpening can lead to the formation of halo's as well, and those can be very visible (an extreme case of this is the clear band around dark objects you often see in a certain kind of HDR images)

Remco

[herbert]

Hi Frank,

When people talk about image sharpness they are usually talking about edges, i.e. when one object stops and another starts. For example a dark grey square inside a light grey square should have a perfectly straight line defining the edge of the inner square. In a perfect world you could take a photo of this and your camera sensor would record the exact colour for the inner and outer squares all the way up the edge where they join. However cameras are not perfect and what happens is that the colours get a bit blended at the edges boundary. This can be due to anti-aliasing on the sensor array, camera shake, lens softness or aberrations, focussing issues, diffraction, etc.

The idea of sharpening is to enhance gradient changes in neighbour pixel values to crisp up edges. This basically adjusts pixels values to make neighbour pixels a bit darker on the dark side of the edge and a bit lighter on the light side of the edge. The effect is flattened back to zero further from the edge (or where there are no local gradients).

In my example above you would make the pixels just inside the dark grey square blacker and the pixels just outside the dark grey square, i.e. on the light grey square, whiter. To get more of a detectable edge you can increase this effect. However there comes a time when the brightening on one side of the edge leads to glowing 'halos' of lighter pixels around objects. This is what most people are talking about when they mention over-sharpening.

To see the adjustment curves I am trying to describe look at this tutorial:

https://www.cambridgeincolour.com/tu...sharpening.htm

It even shows an example of over-sharpening. This is easiest to see in the bottom right corner of the image where the dark head has a light glow around it.

Over-sharpening is easy to spot when you are trying to look for it. Remember that our eyes see an edge as one colour and then another. We do not see halos around objects to make them stand out and so if this is obvious it looks wrong.

Over-sharpening is usually found on images that someone has tried to save with excess sharpening because although it was blurred it framed a nice moment. A better option in this case is to print the image smaller or artistically develop the blur. It is also commonly seen in bad HDR images because the tone mapping operations greatly enhance local contrast and can produce halos around nearly everything in the image.

Sharpening is necessary because of the camera anti-aliasing filter (which partly blurs light falling on neighbour pixels) so it will always be present to some extent. As long as you cannot see it at the desired viewing distance then it is OK. However over-sharpening can be far more of an eye sore than a blurred photo.

It also explains why Colin and others repeatedly point out the need to sharpen at different stages of the workflow:

• Capture sharpening, i.e. when you first import your raw image to fix the anti-aliasing (if you shot JPGs then this is done in camera)
• Artistic sharpening, i.e. to enhance parts of the image
• Output sharpening, i.e. only after you have resized the image (since this blurs edges due to interpolation of pixel values)

Note that a lens cannot be described as over-sharp, only post-processed photos. You definitely want sharp lenses.

Alex

[CBImages]

Hi Frank

I have just finished reading a book all about sharpening and I am still trying to make sense of what I have read, I think another few readings will be undertaken very soon. What I have understood is this:

Our computer screens are trying to display a high res images on low res device.
Our images are usually around 300 dpi and our screens are only around 72 - 102 dpi (a more accurate terminology would be ppi). So some form of interpolation is being carried out to reduce our 300 dpi image down to 72 ppi to allow us to view it at 100% or the 'Fit Screen' mode.
It would therefore be better to view our 300 dpi images at around 30% to get a truer idea of how they will look when printed.
An image that looks slightly over sharpened on a computer screen will in fact look well sharpened when printed. (this is obviously a matter for personal taste).

Alex hits a lot of points on the head in his reply.
1) Sharpening is a matter of taste.
2) Sharpening increases the contrast between the edges in an image by darkening one side and lightening the other, done too much and the darker side becomes black and the lighter side becomes white which causes the halo's. The halo effect can be minimized by using the right sharpening settings - high frequency images (ones with lots of fine detail) need a 'radius' setting of less than 1, usually as low as 0.6. The finer the detail the lower the radius (usually). Low frequency images (ones with smoother graduations in the edge detail) require a higher radius setting, sometimes as high as 1.6 or slightly more. The actual amount is not based on an exact formula but on the eye of the photographer.

3) The 3 stage sharpening process is a valid workflow.
Why go to the lengths of carefully composing an image, accurately metering the scene for the right exposure, then carefully processing the image only to apply 1 level of sharpening?
a) Capture sharpening is done to negate the effects (as Alex states) of the anti aliasing filter which sits in front of our camera's sensor, (It is there to prevent moire patterns and does so by slightly blurring the image). Capture sharpening also corrects the loss of sharpness that occurs through using noise reduction software.

b) Creative Sharpening - is used to enhance part/s of your image, usually the main focal point/s.

c) The final output sharpening is done on the when the image has been resized for its intended output (print or projected display). This is to counteract the softness created during the processing and resizing of our image and also to counteract the softening imparted by our printer as it lays down microscopic dots on the paper.

There is much more I can write but
1) I would only confuse myself and you.
2) My brain hurts already.

Since reading the book my views on sharpening have changed completely, I have employed the 3 step sharpening workflow and I do see a difference in my images. Whether the difference is big enough to persuade others to head down this route is debatable but I am a sad chap and never content to know that 'something works' I need to know 'why it works'.
Sharpening will never rescue a bad image but it can elevate a good image into an impressive work of art.

I am sure Colin will have something to say on the matter (when he gets up ) and his words of wisdom are worth listening too.

Regards

Chris

[arith]

It doesn't happen with film; it is demosaicing that causes the problem. Or what else can be going on that causes digital to become unsharp.

If you add a high contrast edge; is that really what it looked like?

I've for some time relied on deconvolution, I'm told is mathematical reversal of demosaicing. Don't know but the latest program I tried 'InFocus' which did really get back to film quality when used in conjunction with another really expensive sharpening tool; 'Topaz Detail'.

Never use an unsharp mask because it just doesn't look right to me, sometimes use an high pass filter but not often, since that is like a high frequency unsharp mask.

[Colin Southern]

Hi Frank,

It's a complicated business, but as a rule, we need to be doing 3 sharpening passes in our workflow:

Capture Sharpening to counter the effect of the digitization / demosaicing process & anti-aliasing filter. In most cases this will be a Unsharp Mask in the order of 300% @ 0.3 pixel, and 0 threshold (at base ISO) (higher thresholds at higher ISO settings). You'll only see the effect of capture sharpening when looking at the image at 100% though.

Content / Creative sharpening - typically a USM of around 40% @ 4 pixels, but it varies - I like to evaluate this using a much lower resolution, with typically the whole image displayed on the screen at once.

Both of the above are done on the full (uncropped) image (or only lightly cropped).

Output sharpening is done on an image after it's down-sampled for display, or being prepped for printing. For online display I typically down-sample to around 1000 pixels along the longest edge using a bicubic algorithm (not bucubic sharper as one might first expect), and then apply a USM of around 50 to 100% @ 0.3 pixels.

Personally, I don't usually do output sharpening for printing because I'm normally printing either 22 x 44" or 15 x 22", and I just don't need it on canvas.

Over-sharpening can show is a number of ways - in this example I deliberately over-sharpened for effect ...



Click on the image to display it full size, and then look around the railings. This is what's called "low frequency" over-sharpening.

High-frequency over-sharpening is typically seen with fine detail and it results in that portion of an image looking "frosty" - it takes a bit of experience to consciously see it, but when one does, it's pretty easy to fix - just put a gaussean blur over the affected area.

The last area of over-sharpening that I typically see is in images "spat out" (good choice of words!) by many HDR programs where it over-emphasises the transition between contrasty areas (after all, sharpening is all about edge contrast).

Does this help?

[Dave Humphries]

Please allow me, if you will, a small diversion - I promise I will get back on topic soon

Our computer screens are trying to display a high res images on low res device.
Our images are usually around 300 dpi and our screens are only around 72 - 102 dpi (a more accurate terminology would be ppi). So some form of interpolation is being carried out to reduce our 300 dpi image down to 72 ppi to allow us to view it at 100% or the 'Fit Screen' mode.
It would therefore be better to view our 300 dpi images at around 30% to get a truer idea of how they will look when printed.
An image that looks slightly over sharpened on a computer screen will in fact look well sharpened when printed. (this is obviously a matter for personal taste).

All this talk of dpi and ppi will drive you round the bend, please, PLEASE, everyone, do yourselves an enormous favour and, especially when you're working for screen display, just work in PIXELS (the choice is usually there above/below CM and INCHES in the 'Units' drop down and forget dpi/ppi.

You just made your life simpler! Congratulations

Now think like this;
"what size do I want people to see this on CiC?"
You are aiming for them to see it 1:1 *, so down size to 700px on longest edge if you don't want the people who don't know about the Lytebox to see it properly, that size will be good, if a little small, on all screens.

If you have an image that you feel needs to be seen bigger, as I often do, you'll have to host it at TinyPic, or in your own web space.
"So what size should I display it?" you now ask yourself; well, most monitors these days are 1050 or 1080px high, so your image should be no taller than say 1000px, even if it is a portrait orientation shot.
If it is landscape orientation, still don't break the 1000px high rule, but you can take the width out to an average LCD resolution, say upto 1600px max., depending on the aspect ratio of your image.

Now all that said, there are times when, for special reasons, we may want to break those rules, but for the majority of times you will just want an image to be displayable on a typical monitor at 1:1 (so it looks sharp). You don't want people to see a soft version (fit to screen), or be scrolling up and down a huge portrait unable to see chin and eyes on screen together at full resolution.

I said this was mainly for screen didn't I, but I do it this way for print too, the final image size I end up with after PP cropping will, when divided by 100 or 200 (easy enough to do in your head) tell me how good the print will be. For example; 4000/100 is 40 inches at 100 ppi or 20 inches at 200 ppi, now being pragmatic, that's all I need to know. When I see the angst some people get into, and the knots they tie themselves in "changing resolution" to make their print good enough, I feel sorry for them, it is so unnecessary.

OK, I promised to go back on topic, so here we go ...

After you have downsized (for screen only), you really should (read "must") sharpen again.
In most of cases I do this with USM at;
A 0.3px radius (guaranteeing no halos will be visible) and,
An amount of 70 - 120%, depending on lots of things; primarily how much of any creative sharpening is still evident, not much if you down sized from 4 or 5 thousand pixels.
Use as low a Threshold as you can get away with, again, if the downsize was significant, that should have reduced most of the noise, allowing a threshold of 0, 1 or 2, but if the image had a significant crop in PP, hence little downsize was applied, you may need this to be higher (e.g. 3 - 9).

Things that make a shot look over sharpened are (IMO);
An obvious hard edged halo of 1 or 2 pixels (too wide a radius in final/output sharpening, or sharpening during the downsize (don't use "bi-cubic sharper"!))
Speckly fine detail (too high an amount, usually with slightly too wide a radius aswell)
Gritty noise (too low a threshold, or too much noise due to crop and/or high iso noise not being removed in PP)

Hope that helps,

[Shadowman]

I've seen the oversharpening issue mentioned enough times that I am starting to wonder exactly what it means. Is oversharpened a personal taste or a definable point which anyone could identify by specific characteristics in the image?

When two people look at an image and one says that it is oversharpened and the other says that it is not, is there a way to define the truth? Or is it so subjective that we just shrug our shoulders and agree with one or the other?

The oversharpening issue depends on how the image is viewed. If you zoom into an image then you can see more clearly the effects of sharpening. I often see it when doing PP and I usually back off if it appears that the edges have become frayed or if I begin to see noise around the edges of a subject.

[FrankMi]

I want to thank Willie, Remco, Alex, Chris, Steve, Colin, Dave and John for taking the time to help me understand not only how I can tell if an image has been over sharpened and why, but also for sharing some guidelines for preventing this from occurring in the first place.

From what I now understand, the appearance of halos at the border of the light and dark parts of an image is one way to determine that the image has been over sharpened. Two other indicators are noise specks in the fine detail and gritty noise ( I think Dave is referring to the dark portions of the image).

I can clearly see the halo in the example of capture sharpening when the radius is above 1 pixel in the tutorial link provided by Alex https://www.cambridgeincolour.com/tut...sharpening.htm. This gives me a rational starting point for pixel size when doing Capture Sharpening.

HDR Tonemapping is particularly sensitive to producing halos and now I understand why.

Chris points out that images that already have relatively sharp detail may suffer from halos if the sharpening radius is above 1 pixel and may need to be much lower, whereas images with a more gradual transition from light to dark would require a larger pixel size, closer to 1.6 pixels to sharpen effectively. This helps explain why artistic sharpening may need to be applied differently to specific areas of an image.

Changing the pixels per inch to either reduce the image size for display on the web or increase the image size for printing gives us the same type of issues as the Capture process creates .

If I have it correct, the consensus for the starting points for using the unsharp mask are as follows:

Capture Sharpening – Amount, 300%; Radius, 0.3 pixels; Threshold, 0

Artistic (or Creative) Sharpening – Amount, 40%, Radius, 4 pixels, Threshold, 0

Output Sharpening – Amount, 50-120%; Radius, 0.3 pixels; Threshold, usually less than 3 but not over 9

Although a number of factors can change these starting settings, you don’t want to vary too far from these starting points unless you really understand why.

Please let me know if I missed or misapplied any of the important points. Thanks guys, this should make the sharpening process easier to implement correctly.

[Colin Southern]

Changing the pixels per inch to either reduce the image size for display on the web or increase the image size for printing gives us the same type of issues as the Capture process creates .

PPI confuses many many people. Ultimately, it's just a number (number of pixels along one dimension divided by the length of that dimension). It doesn't have any effect on the size of an image, as - ultimately - an image prints of displays at the size it's told to - it DOES however (potentailly) have a bearing on the QUALITY of an image at a given size. As Dave says, "just forget about it".

I had an example in a thread somewhere where I posted the same image at different PPI to prove the point, but I can't find it now (if anyone knows where it is, I'd be grateful if they could post a link to it).

The following link has a bit of info that might help though.

New here and to Photoshop and need to print large sizes.