A technical view of over-sharpening

[xpatUSA]

I have QuickMTF, a software which can analyze a slanted straight edge and provide MTF and edge-response graphs for comparison.

While looking at different ways to de-Bayer a raw file, I tried the simple method of converting RGGB quads to RGB pixels (dcraw -h -T). The resulting image lacked contrast and was slightly soft so I had at it in FastStone Viewer with Levels followed by three stages of sharpening at 1.3 then 0.7 then 0.3px.

Voila:



Obviously well over the top!

Comments/questions invited ...

[Manfred M]

I have QuickMTF, a software which can analyze a slanted straight edge and provide MTF and edge-response graphs for comparison.

While looking at different ways to de-Bayer a raw file, I tried the simple method of converting RGGB quads to RGB pixels (dcraw -h -T). The resulting image lacked contrast and was slightly soft so I had at it in FastStone Viewer with Levels followed by three stages of sharpening at 1.3 then 0.7 then 0.3px.

Voila:



Obviously well over the top!

Comments/questions invited ...

Interesting but not terribly useful without the full-size corresponding images.

[billtils]

I started a reply with "If I knew/understood what you were doing I'm sure I'd appreciate it But I don't.". However, your post made me go and do some reading - thanks Dan, I'm a million miles away from "expert" but a whole lot more aware, and ready to read some more.

[Acorn]

My judgement of over-sharpening is based purely along visual lines. If there's a fine halo to outlines it's a sure sign of over-sharpening.

[xpatUSA]

Interesting but not terribly useful without the full-size corresponding images.

[xpatUSA]

My judgement of over-sharpening is based purely along visual lines.

Prompted by Manfred, I've just posted the analyzed images FYI.

If there's a fine halo to outlines it's a sure sign of over-sharpening.

In the posted edge response graphs, please see the gray bar above each graph. Above the over-sharpened graph, the approx 2px halo is clearly shown ...

[xpatUSA]

I started a reply with "If I knew/understood what you were doing I'm sure I'd appreciate it But I don't.". However, your post made me go and do some reading - thanks Dan, I'm a million miles away from "expert" but a whole lot more aware, and ready to read some more.

The things to note in particular about the over-sharpened graphs are A) the huge overshoot peak in the edge response and the far too low 10-90pct rise number **, and B) the MTF should never go over a value of one - because that means the camera is applying more contrast than the original scene had, which ain't right (think lens MTF numbers, never greater than 100%). In other words, the MTF50 and the MTF50P numbers should not be different.

** 0.81 is way too steep, 1.25 is about right, 2 to 3 is getting a bit soft ...

[Manfred M]

An interesting analysis Ted, although I question the "real world" relevance of the analysis.

The two things that were stressed when this subject came up when this subject was discussed in a number of photography courses I took were:

1. Size matters - all final sharpening is applied when the image has been resized to the final viewing size, so this is an important parameter that needs to be considered. This means that sharpening can be considered to be device specific; the same image viewed on a phone can look fine while it will look over-sharpened on a large screen. When it comes to prints, viewing distance becomes another factor, so a large panorama can use more sharpening than a small format print; and

2. All sharpening is really is how boundaries between the dark and light parts of the image are handled, so this very much has a "personal taste" factor. When the halos become distracting, the image is generally viewed as over-sharpened. Parts or all of an image can look "too crunchy" and still not have any halos; my view is that this is an oversharpening issue too.

[xpatUSA]

An interesting analysis Ted, although I question the "real world" relevance of the analysis.

<>

<>

Well, I understand the oft-mentioned Real World to include testing of cameras/lenses/images for sharpness and other properties.

For example, is it the opinion here that lens MTF curves are relevant to the Real World or not?

Or, do people in the Real World buy lenses without looking at technical curves found in reviews?

[Manfred M]

For example, are lens MTF curves relevant to the Real World, I wonder.

I don't care how good the resolution of the lens is, if the photographer has missed focus or has camera shake. These metrics are important as the set the ceiling on image quality. Put a human behind the stunning lens and he (or she) will likely not take full advantage of it.

Those are the reasons I try to use a heavy-duty tripod as often as practical or use flash (especially studio flash) as well as shooting tethered. That helps me nail the technical aspects of quality.

I've received awards taken with cameras and lenses that would hardly be considered "high end" when I look that the test results on sites like DxOMark.

https://www.dxomark.com/Lenses/Nikko...n-D750---__975

A person who knows what they are doing can pull off great images, regardless of camera / lens quality or lack thereof with modern gear.

[xpatUSA]

Originally Posted by xpatUSA
For example, are lens MTF curves relevant to the Real World, I wonder.

I don't care how good the resolution of the lens is, if the photographer has missed focus or has camera shake. These metrics are important as the set the ceiling on image quality. Put a human behind the stunning lens and he (or she) will likely not take full advantage of it.

Manfred, by diverting attention from lens quality to the photographer's ability, a classic Straw Man argument has been introduced.

Those are the reasons I try to use a heavy-duty tripod as often as practical or use flash (especially studio flash) as well as shooting tethered. That helps me nail the technical aspects of quality.

I've received awards taken with cameras and lenses that would hardly be considered "high end" when I look that the test results on sites like DxOMark.

https://www.dxomark.com/Lenses/Nikko...n-D750---__975

A person who knows what they are doing can pull off great images, regardless of camera / lens quality or lack thereof with modern gear.

I'm not sure that the above is a valid rebuttal to my claim that lens testing is part of the Real World.

[xpatUSA]

Staying in the Real World a photographer decides to compare de-mosaicing algorithms applied to the same raw file.

On this occasion, he decides to compare 'AHD' with the most simple of all algorithms where each RGGB 'quad' is converted to one RGB pixel. A reasonable Real World activity, I claim.

dcraw AHD:


dcraw -h


I suppose one could compare these images by eye perhaps by flipping back and betwixt the two.

Or, one could get numbers from analytical software, voila:



I claim that using the software and getting numbers is a Real World activity - no difference than comparing colors with the use of a color-picker and just as relevant to the task in hand.

[Manfred M]

Manfred, by diverting attention from lens quality to the photographer's ability, a classic Straw Man argument has been introduced.

I guess I always look at things from a system, rather than component standpoint. The final product (in this case an image) is always limited by its weakest link. When it comes to making an image (which has technical, compositional and impact) connotations, which is the most important? In my view, the technical performance of the equipment is usually the least important element that goes into a great image.

The problem with looking at an MFT chart is that it tells us nothing about what we are being shown. Is it the theoretical design data that may or may not be reproduced in the "real world"? Is it based on actual test data on one or more production lenses? Is it test data from a "cherry picked" lens selected so that the numbers look good? Manufacturing variances and tolerances never seem to be a consideration when people write about gear.

[Manfred M]

I claim that using the software and getting numbers is a Real World activity - no difference than comparing colors with the use of a color-picker and just as relevant to the task in hand.

I would suggest these are more akin to lab tests than anything to do with the real world.

I suspect that test charts are some of the least photographed items around.

[xpatUSA]

In view of the generally negative reception of this thread by the cognoscenti, I retire from this discussion.

Stay well ...