The dangers of pixel peeping

[bjanes]

Michael's recent megapixel shootout has provided us with a lot of useful information and the raw files are available for further analysis. When comparing perceived sharpness between two cameras, one method is to view the file in Photoshop at a very large size (up to 400%) and compare the rendering of fine details, such as the engraving lines on the currency. In the context of MTF analysis, this shows resolution at high frequency and relatively low contrast. However, perceived sharpenss in the print viewed at normal distances is more closely related to MTF at lower frequencies. For a dramatic example, view this post:

http://www.wellsresearch.com/technical_lib...ons_learned.htm

The picture on the left looks better to most observers, but the picture on the right has more resolution when the USAF test target is view at high magnification in the picture on the right. Part two of the analysis, gives another example. One could argue that the examples are not that realistic, since high resolution and low contrast are easily seen on the USAF target at high magnification.

Studies at Kodak by Ed Grainger have shown that subjective quality Factor (SQF) is the average MTF in a photo from 0.5 to 2 cycles per mm when view at a normal distance (13.5 inches). For an 8 by 12 inch print, the image is enlarged 8x with a 35mm negative or full frame 35mm type sensor, and the corresponding critical resolutions on the film or sensor would be 4 to 16 cycles/mm. The above print resolution (0.5 to 2 cycles per mm) is far less than the eye can resolve; an individual with 20/20 vison can resolve about 10 cycles/mm under these conditions (cycles/mm = line pairs/mm).

Bob Atkins provides additonal information on his web site and considers larger pictures where the critical MTFs must be shifted upward. He is apparently associated with Popular Photography magazine, but I wouldn't necessarily hold this against him.  

http://bobatkins.com/photography/technical/mtf/mtf4.html

If you use Norman Koren's Imitest program, you can perform MTF analysis with your own camera or download pictures taken by others for analysis. One good source is from the DPReview tests. I downloaded the results for the Canon EOS 1Ds Mark II and plotted them with the x-axis as a log scale as suggested. The MTF is shown unsharpened and with standardized sharpening. Since the log scale is hard to read, the triangles towards the bottom shows the 4, 8, 12, 16, 32, 48, and 64 lp/mm values.

[attachment=808:attachment]

Here is an an analysis for an an 8 by 12 inch print, magnification = 8, with and without sharpening. SQF is represented by the colored areas under the curve.
[attachment=809:attachment]

and for a 24 by 36 inch print, mag = 24
[attachment=810:attachment]

The graphs show that for an 8 by 12 inch print, the camera gives nearly maximal SQF. For a 24x enlargement, SQF with sharpening is still very good. At larger enlargements, SQF falls and one would have to use a medium format back or a scanning back such as the Better Light for optimum results. Popular Photography claims to use SQF in their tests, but this is questioned by David Jacobson in his lens FAQ:

http://www.photo.net/photo/optics/lensFAQ

The point of all this is that, in additon to viewing the test photo on the screen at high magnification in order to check MTF at high frequency (which is good, of course, but not the primary determinant of percieved sharpness), one must also view the the photo at lower resolution in order to get some impression of the MTF at the critical lower frequencies. I do not know what screen resolution would be best for this purpose, but it seems to me like one should start with a resolution similar to the size of the intended print. Of course, it would best to view an actual print or at least a cropped area of the print at the same resolution.

[Ray]

The point of all this is that, in additon to viewing the test photo on the screen at high magnification in order to check MTF at high frequency (which is good, of course, but not the primary determinant of percieved sharpness), one must also view the the photo at lower resolution in order to get some impression of the MTF at the critical lower frequencies. I do not know what screen resolution would be best for this purpose, but it seems to me like one should start with a resolution similar to the size of the intended print. Of course, it would best to view an actual print or at least a cropped area of the print at the same resolution.
[a href=\"index.php?act=findpost&pid=70268\"][{POST_SNAPBACK}][/a]

The point also is that these SQF tests you refer to were carried out in the 1970's before Photoshop and digital cameras for the general public were available. That does not mean of course that the 'subjective quality factor' is no longer valid. However, the methods of achieving that SQF have progressed and that factor may now no longer be so dependent upon lens contrast at low frequencies.

In the example of the yellow cars in your first link showing the lower resolution image appearing to be sharper than the higher resolution image, the higher resolution, slightly blurry looking car can be made to 'appear' as sharp as the more contrasty car by simply applying a bit of sharpening. (I tried it, using a blur width of 3 pixels in Focus Magic at 100%).

The real question here is, in what ways can a contrasty lens produce an appearance of sharpness that cannot be created or imitated in Photoshop with an appropriate amount and type of sharpening, including edge sharpening and local contrast enhancement?

The bottom line might well be, no amount of processing can create detail where none existed, but processing can create an appearance of additional sharpness or accutance (SQF) to compensate for a lack of contrast of the lens at low frequencies.

It would be interesting to see some comparisons between a first rate prime and a first rate zoom, (or 2nd rate prime), with the image from the less contrasty lens processed more to mimic that appearance of greater accutance.

Also, it's always been the case that the smaller the print, the lower the image resolution required because the eye cannot discern (apparently) a greater resolution than 4 to 6 lp/mm from a 'normal' viewing distance. I recall somewhere, you quoted a very precise figure of 6.7 lp/mm. However, if we take the figure at the top of the range, 6 lp/mm, and do a bit of simple maths, then we cannot avoid a conclusion that no current Canon camera, except possibly the 1Ds2, is able to take an image of sufficiently high resolution that would meet this very high standard on an 8x12" print.

For example, 6 lp/mm across the width of a vertically oriented 8x12' print requires a print resolution of 6x200=1200 line pairs.

The 20D and 30D have a maximum resolution of around 60 lp/mm. 15x60=900 line pairs. That's a big shortfall. According to Dpreview, the 5D has a 'total picture height' number of lines that varies in the vertical and horizontal direction. Why this should be so, I have no idea, but the higher figure of 2300 lines does not quite meet the standard of 1200 line pairs, but that's close enough for me. The lower figure of 2000 lines is a bit too far off.

My general view is, I do not always know when a shot I take might look impressive blown up larger at a later date, for a customer perhaps or just because I might later be able to afford a larger printer. There's no disadvantage in capturing an image which has higher resolution for one's immediate purposes, but the reverse  certainly does have some disadvantages, with the possible exception of DoF appearances on print, the smaller print producing greater DoF, which might not be the desired effect.

[bjanes]

The real question here is, in what ways can a contrasty lens produce an appearance of sharpness that cannot be created or imitated in Photoshop with an appropriate amount and type of sharpening, including edge sharpening and local contrast enhancement?

The bottom line might well be, no amount of processing can create detail where none existed, but processing can create an appearance of additional sharpness or accutance (SQF) to compensate for a lack of contrast of the lens at low frequencies.

It would be interesting to see some comparisons between a first rate prime and a first rate zoom, (or 2nd rate prime), with the image from the less contrasty lens processed more to mimic that appearance of greater accutance.
[{POST_SNAPBACK}][/a]

Ray,

Thanks for responding to my post--you are the first to do so, and there does not appear to be a lot of interest in this type of analysis.

I think there are limitations in the use of sharpening to boost MTF at low frequencies. If you use a radius large enough to improve edge sharpness at low frequency, then the higher frequecies will be blurred. Of course, you can use edge masks and other techniques, but there are still limitations. Nothing looks worse than an oversharpened image.

As to the applicability of SQF in the digital area, you raise a good point. Intuitively, I would think that MTF on the print is what matters and not how it is achieved with analog techniques or digitally.

If you want to get into Fourier transforms, point spread functions, etc, Norman Koren has an analysis of sharpening on his web site. The details of this type of analysis are beyond my understanding of the math, but the basic concepts are not that difficult to grasp.

[a href=\"http://www.imatest.com/docs/sharpening.html]http://www.imatest.com/docs/sharpening.html[/url]

Also, it's always been the case that the smaller the print, the lower the image resolution required because the eye cannot discern (apparently) a greater resolution than 4 to 6 lp/mm from a 'normal' viewing distance. I recall somewhere, you quoted a very precise figure of 6.7 lp/mm. However, if we take the figure at the top of the range, 6 lp/mm, and do a bit of simple maths, then we cannot avoid a conclusion that no current Canon camera, except possibly the 1Ds2, is able to take an image of sufficiently high resolution that would meet this very high standard on an 8x12" print.

For example, 6 lp/mm across the width of a vertically oriented 8x12' print requires a print resolution of 6x200=1200 line pairs.

The 20D and 30D have a maximum resolution of around 60 lp/mm. 15x60=900 line pairs. That's a big shortfall. According to Dpreview, the 5D has a 'total picture height' number of lines that varies in the vertical and horizontal direction. Why this should be so, I have no idea, but the higher figure of 2300 lines does not quite meet the standard of 1200 line pairs, but that's close enough for me. The lower figure of 2000 lines is a bit too far off.
[{POST_SNAPBACK}][/a]

The precise figure for the resolution of the eye that I stated previously should have been 6.9 lp/mm, which is taken from Norman's web site. He calculated 6.88, but I rounded it off to 6.9. I think this derives from a calculation that the eye can resolve approximately 1 minute of arc.

The differences in horizontal and vertical resolution are even more prominent in Nikon cameras such as the D70, and I have read that the pixels are rectangular and not square. There may also be assymetry in the AA filters and microlenses.

The whole point of the analysis was that, for a print viewed at 34 cm (13.5 inches), the important MTFs for perceived image sharpness are between 0.5 and 2 lp/mm. For an 8 by 10 (or 8 by 12) inch print, this corresponds to a resolution of 4 to 16 lp/mm on the sensor of a full frame camera. For practical purposes, it is not necessary to achieve the higher resolution that you quote. Of course, there is no harm with more resolution if MTF in the critical range is not compromised. You will then be able to print at larger sizes.

[a href=\"http://www.normankoren.com/Tutorials/MTF.html]http://www.normankoren.com/Tutorials/MTF.html[/url]

Unfortunately, I do not have the 1Ds M2, but the SQF analysis predicts that a 24 by 36 inch print is about max for good perceived print quality. How does this compare with results achieved by those who do have the camera?

Ther corollary of the above is that the important MTFs are between 12 and 48 lp/mm for a picture of this size, and we should be looking at details in this range, not barely perceived low contrast detail at greater magnification as is done with the pixel peeping.

Bill

[Kenneth Sky]

when you view a print larger than 8x10 (12) don't you step back proportionately? so if the eye resolves 6 lines at 13.5 in. won't it remain the same?

[Geoff Wittig]

The 24x36" limit seems about right to me. I own a 1Ds II, and I have made a fair number of prints this size. With perfect technique (solid tripod, mirror lock-up, careful focus, sharp L-series glass and ISO 100) and judicious sharpening, even images with lots of small detail like forest scenes can look terrific. On the other hand, minor image flaws become painfully obvious.

[bjanes]

when you view a print larger than 8x10 (12) don't you step back proportionately? so if the eye resolves 6 lines at 13.5 in. won't it remain the same?
[a href=\"index.php?act=findpost&pid=70692\"][{POST_SNAPBACK}][/a]

Yes, one normally views a large print from a greater distance. However, early on in this thread some perfectionists stated that they wanted to examine the prints up close. There is a certain satisfaction in having prints that can withstand this scrutiny.

[Ray]

The whole point of the analysis was that, for a print viewed at 34 cm (13.5 inches), the important MTFs for perceived image sharpness are between 0.5 and 2 lp/mm. For an 8 by 10 (or 8 by 12) inch print, this corresponds to a resolution of 4 to 16 lp/mm on the sensor of a full frame camera. For practical purposes, it is not necessary to achieve the higher resolution that you quote. Of course, there is no harm with more resolution if MTF in the critical range is not compromised. You will then be able to print at larger sizes.

[a href=\"index.php?act=findpost&pid=70674\"][{POST_SNAPBACK}][/a]

Bill,
Norman Koren also makes the following observatrion:

This means that the MTF of a 35mm imaging system (lens + film, etc.) at and above 55 lp/mm has little effect on the appearance of an 8x10 inch print enlarged 8x. That's why the highest spatial frequencies used for lens tests is 40 lp/mm, which provides an excellent indication of a lens's perceived sharpness in an 8x10 inch print enlarged 8x. Of course higher spatial frequencies are of interest for larger prints.

I'm in the business of making larger prints. I have an Epson 7600 printer. I recently framed the following 22x34" print, from a 5D image, and placed it in a prominent position above the TV set and bookshelf. It would not be possible to view this print from closer than the diagonal. All visitors, after expressing the usual exclamations of praise, wow!, great!, amazing!, just like a painting! etc etc, always at some point view the photo from as close as they can get.

[attachment=820:attachment]

ps. Sigma 15-30 zoom at 15mm. No cropping.

[bjanes]

Bill,
Norman Koren also makes the following observatrion:

This means that the MTF of a 35mm imaging system (lens + film, etc.) at and above 55 lp/mm has little effect on the appearance of an 8x10 inch print enlarged 8x. That's why the highest spatial frequencies used for lens tests is 40 lp/mm, which provides an excellent indication of a lens's perceived sharpness in an 8x10 inch print enlarged 8x. Of course higher spatial frequencies are of interest for larger prints.

I'm in the business of making larger prints. I have an Epson 7600 printer. I recently framed the following 22x34" print, from a 5D image, and placed it in a prominent position above the TV set and bookshelf. It would not be possible to view this print from closer than the diagonal. All visitors, after expressing the usual exclamations of praise, wow!, great!, amazing!, just like a painting! etc etc, always at some point view the photo from as close as they can get.

ps. Sigma 15-30 zoom at 15mm. No cropping.
[a href=\"index.php?act=findpost&pid=70738\"][{POST_SNAPBACK}][/a]

Raw,

A very nice picture, I must say. It's unfortunate we can't see it at full resolution.

Yes, I had read that statement by Norman. He is talking about the maximum resolving power of the eye, not the critical resolutions used in the SQF analysis. He was using 6.88 lp/mm as the resolution of the eye, which would require 55 lp/mm on the film or sensor for the 8x print. He is using a higher standard of sharpness than the SQF analysis uses.

For an 8 by 10 inch print (M = the critical SQF resolutions of 0.5 to 2 lp/mm on the print correspond to 4 to 16 lp/mm on the full frame sensor. However, for your print where M = 22, the corresponding resolutions on the sensor are 11 to 44 lp/mm. To reach the limit of the eye's resolution at M = 22 on the print would require 150 lp/mm on your sensor, which is well beyond its capability. However, the sensor is capable of achieving the 44 lp/mm critical resolution required by the SQF analysis. Your example is a validation of the SQF analsysis: the image appears critically sharp as predicted. If 150 lp/mm were required for critical sharpness, you would be in trouble.

If you look at my SQF graph for the EOS 1DsM2 for a 24 by 36 inch print, you will see that MTF has fallen off dramatically at 48 lp/mm (which is where the colored region ends on the x-axis) even with sharpening, and the area under the curve is less than one. SQF has begun to suffer, but it is still good. Further enlargement would not produce optimal results.

[Gary Ferguson]

A very interesting post. But thinking about individual lenses isn't it true that, generally speaking, a lens that delivers the goods at the pixel peeping level of absolute resolution will also tend to be the strongest performer when it comes to the high contrast that drives SQF?

Up until the 60's or 70's the orthodox view was that some lens manufacturers (Zeiss is often quoted) favoured high contrast for larger image structures over absolute resolution, where as other manufacturers (Leica for example) promoted resolution above "contrast". With the growing awareness of MTF testing in the 80's and 90's this view seemed to fade, being replaced with the opinion that any lens that delivers high resolution will also tend to deliver higher contrast across the lower frequencies. And that's the impression I'm left with after looking at the MTF charts for many lenses, off-hand the only exception I can think of to this general rule is the Hasselblad 100mm lens, which scores relatively poorly at 10 lppm given its relatively high result at 40 lppm.

However, wouldn't pixel peeping the finest details also usually serve to identify those optics that produce the impactful and crisp results required under the SQF approach?

Incidentally, I also use a Canon 1Ds MkII. If I print really large I can't unequivocably state that the 1Ds MkII is up to the job. Even with meticulous technique sometimes it produces large satisfying images, but sometimes it doesn't. The key determinant of success seems to be the nature of the subject rather than the pixel count or lens quality.

Most of my work is architectural, as a rough rule of thumb a modern building consisting of relatively large, graphical shapes will tend to enlarge very well. On the other hand older buildings that are rich in textural detail don't usually hold up easily to giant enlargement. I suspect that our brains examine images with a ready made check list, so for architectural shots we expect the half-timbering of an Elizabethan structure to show the fissured texture of ancient wood, and if we can't see this detail we're disappointed. Likewise we know a dark blue-grey roof is likely to consist of slate, and therefore we expect to see the individual slate tiles clearly delineated, and if they're not we tend to judge the shot as slightly soft.

The converse of this is that macro shots seem often to be applauded as extremely sharp when under more objective scrutiny it's clear that they're anything but. Maybe seeing a photograph of the anatomy of a fly's eye is so powerful and striking that we automatically consider it razor sharp even if it's a bit soft.

In other words I suspect that there's pyschological considerations, based on our expectations of seeing certain levels of detail, that shape our response to an image and our judgement of its relative sharpness.

[Ray]

In other words I suspect that there's pyschological considerations, based on our expectations of seeing certain levels of detail, that shape our response to an image and our judgement of its relative sharpness.
[a href=\"index.php?act=findpost&pid=70804\"][{POST_SNAPBACK}][/a]

Good points, Gary. They make sense to me. There's also another issue relating to the variability of peoples' eyesight. I'm fairly relaxed about blow-ups which might be larger than some critical photographers would want to make, because I'm long-sighted. My partner, however, is short-sighted. She can read the small print on bottles of suace by taking off her reading glasses and peering closely.

It would be interesting to do the following experiment. Take a wide-angle shot of a detailed scene (like my enlargement above), zoom in on a section about 1/8th the area and take another shot. Compare an 8x12" print of the zoomed shot against the same sized setion of a 24x36 enlargement.

[BernardLanguillier]

I fully agree with this as well.

Foliage is another good test of the ability of a camera to render very fine detail. Images of rocky landscape do in my experience print large much better than those landscapes with trees in the distance.

The eye just goes to look for the rendition of these small leaves etc...

Regards,
Bernard

[bjanes]

A very interesting post. But thinking about individual lenses isn't it true that, generally speaking, a lens that delivers the goods at the pixel peeping level of absolute resolution will also tend to be the strongest performer when it comes to the high contrast that drives SQF?

Up until the 60's or 70's the orthodox view was that some lens manufacturers (Zeiss is often quoted) favoured high contrast for larger image structures over absolute resolution, where as other manufacturers (Leica for example) promoted resolution above "contrast". With the growing awareness of MTF testing in the 80's and 90's this view seemed to fade, being replaced with the opinion that any lens that delivers high resolution will also tend to deliver higher contrast across the lower frequencies. And that's the impression I'm left with after looking at the MTF charts for many lenses, off-hand the only exception I can think of to this general rule is the Hasselblad 100mm lens, which scores relatively poorly at 10 lppm given its relatively high result at 40 lppm.

However, wouldn't pixel peeping the finest details also usually serve to identify those optics that produce the impactful and crisp results required under the SQF approach?
[{POST_SNAPBACK}][/a]

In general a sharp lens is also a contrasty lens as you say, so why not just look at the MTF at high frequencies and why does PhotoDo weight MTF at 10% more than MTF at 40%? Why even bother testing MTF at 10 lp/mm? These MTFs apply to 8 by 10 inch prints observed up close, but the same considerations also apply to larger prints. The required MTFs simply increase across the board. With regard to the Zeiss vs Leitz characteristics, I think that it is the other way around: Zeiss favored resolution while Leitz favored contrast. As Michael points out in his MTF essay, "For the optical designer contrast and resolution are in conflict. Increase one and you reduce the other." For Leitz to favor contrast was sensible considering the classic Leica shooting style involving candid shots without a tripod--you are not getting high resolution under those conditions, so go for contrast. Color adds contrast, so the equation changes for color photography, where more resolution may be of interest.

[a href=\"http://www.luminous-landscape.com/tutorials/understanding-series/understanding-mtf.shtml]http://www.luminous-landscape.com/tutorial...nding-mtf.shtml[/url]

Of course, in the final print, the MTF of the lens, camera, and printing system are all important. MTF above Nyquist results in alaising, which is why digital cameras have AA filters. An ideal lens for digital would have high MTF above Nyquist, but MTF below Nyquist is not only useless but also harmful.

Incidentally, I also use a Canon 1Ds MkII. If I print really large I can't unequivocably state that the 1Ds MkII is up to the job. Even with meticulous technique sometimes it produces large satisfying images, but sometimes it doesn't. The key determinant of success seems to be the nature of the subject rather than the pixel count or lens quality.

Most of my work is architectural, as a rough rule of thumb a modern building consisting of relatively large, graphical shapes will tend to enlarge very well. On the other hand older buildings that are rich in textural detail don't usually hold up easily to giant enlargement. I suspect that our brains examine images with a ready made check list, so for architectural shots we expect the half-timbering of an Elizabethan structure to show the fissured texture of ancient wood, and if we can't see this detail we're disappointed. Likewise we know a dark blue-grey roof is likely to consist of slate, and therefore we expect to see the individual slate tiles clearly delineated, and if they're not we tend to judge the shot as slightly soft.

The converse of this is that macro shots seem often to be applauded as extremely sharp when under more objective scrutiny it's clear that they're anything but. Maybe seeing a photograph of the anatomy of a fly's eye is so powerful and striking that we automatically consider it razor sharp even if it's a bit soft.

In other words I suspect that there's pyschological considerations, based on our expectations of seeing certain levels of detail, that shape our response to an image and our judgement of its relative sharpness.
[a href=\"index.php?act=findpost&pid=70804\"][{POST_SNAPBACK}][/a]

I agree fully with your observations involving the psychology of sharpness and how the subject matter (mostly low frequency objects or also high frequency details) affect perceived sharpness. No one will deny that resolution of fine details (as shown by pixel peeping) is desirable, but one should also pay attention to contrast at lower frequency.

[Ray]

I think there are limitations in the use of sharpening to boost MTF at low frequencies. If you use a radius large enough to improve edge sharpness at low frequency, then the higher frequecies will be blurred.
[a href=\"index.php?act=findpost&pid=70674\"][{POST_SNAPBACK}][/a]

Bill,
There are definitely limitations in the use of sharpening and I agree that oversharpening is not good, but this idea that using a large radius in sharpening (to increase local contrast) might blur the higher frequencies does not seem right to me. At least, that's not the experience I'm getting. Where did you pick up this idea?

Whilst getting my images into some sort of data base, I came across the first test images I took the day I unpacked my brand new 20D. They look so sharp, I thought I might be able to use them to refute this assertion of yours that local contrast enhancement blurs the higher frequencies.

Immediately below is the full image, uncropped. Nothing special, just a female King parrot looking a bit startled. Taken with my 100-400 zoom at 400mm, f8 and 320th second, and ISO 800.

[attachment=822:attachment]

When I zoomed on this image, on screen, to 300%, I noticed that the feathers took on an appearance of test chart lines. How many line pairs per mm do these feathers represent, I thought.

Below are the 400% crops which I hope will demonstrate my point.

[attachment=825:attachment]

Now before we go any further, I'm going to lay bare my mathematical reasoning. If it's flawed, I'm sure Jonathan will jump on me like a tonne of bricks and make me look like a complete twerp. BJL might be more forgiving.

The image as converted is 248mm wide. At the top of the crops we have a scale in mm. When I count the feathers, lines and spaces, I get 13.5 for 2mm. I refuse to exaggerate, so let's call it 13 lines which equal 6.5 line pairs per 2 mm, or 3.25 lp/mm.

The width of the 20D sensor is 15mm. The width of the on-screen image is 248mm. Therefore, the enlargement factor is 248/15=16.5.

16.5x3.25=53.  Those feathers represent an on-sensor resolution of 53 lp/mm. The 20D can manage a few more, but as you should have noticed, those lines are not all equal in width. If they were, I think we would have reached full resolution of 60 lp/mm. (Do I recall hearing from some photographic experts that test chart lines were irrelevant to real world photography?)

The images were converted using RSP with 'detail extraction' at maximum and -10 sharpening. The top crop has no 'local contrast' enhancement. The lower crop had USM at 30% and a pixel radius of 50, applied twice. The 13 lines are still as clear as a bell. I rest my case.

Don't know what theses thumbnails are doing below, but they appear to be edited versions that are still hanging around.

[bjanes]

Bill,
There are definitely limitations in the use of sharpening and I agree that oversharpening is not good, but this idea that using a large radius in sharpening (to increase local contrast) might blur the higher frequencies does not seem right to me. At least, that's not the experience I'm getting. Where did you pick up this idea?
[a href=\"index.php?act=findpost&pid=70849\"][{POST_SNAPBACK}][/a]

Ray,
That idea comes from Fraser & Blatner, Real World PSCS2, page 466 in the discussion of image detail and sharpening radius, where they discuss the proper sharpening radius to use for high and low frequency images. "... an overly large radius can actually wipe out the detail it's susposed to be accentuating. ... The larger radius sharpens the larger elements well, but the more delicate elements are lost."

The images were converted using RSP with 'detail extraction' at maximum and -10 sharpening. The top crop has no 'local contrast' enhancement. The lower crop had USM at 30% and a pixel radius of 50, applied twice. The 13 lines are still as clear as a bell. I rest my case.

[a href=\"index.php?act=findpost&pid=70849\"][{POST_SNAPBACK}][/a]

In your example, I'm not certain if you are observing actual detail in the feathers or merely alaising. At any rate, your choice of a sharpening radius is interesting. To accentuate details in the range of one or two pixels, you should be using a smaller radius and a larger amount. In their example of a high frequency image, the above authors suggest a radius of 0.6 and amount of 275. Your settings are what Dan Margulis calls "hiraloam", which he uses for portraits so as not to accentuate blemishes.

[Ray]

In your example, I'm not certain if you are observing actual detail in the feathers or merely alaising. At any rate, your choice of a sharpening radius is interesting. To accentuate details in the range of one or two pixels, you should be using a smaller radius and a larger amount. In their example of a high frequency image, the above authors suggest a radius of 0.6 and amount of 275. Your settings are what Dan Margulis calls "hiraloam", which he uses for portraits so as not to accentuate blemishes.
[a href=\"index.php?act=findpost&pid=70877\"][{POST_SNAPBACK}][/a]

The detail is not aliasing as you can see in the 100% crop of a patch of feathers below.

[attachment=826:attachment]

The sharpening of this image was done in RSP. The use of USM in Photoshop with a large radius of 50 pixels is not a normal sharpening procedure but a technique known as 'local contrast enhancement'. Michael has a tutorial on it.

I mention this in connection with an earlier point I made, namely that a lack of contrast at low frequencies that might be more critical for subjective sharpness, is no longer such a disadvantage if one happens to have used a high resolution lens as opposed to a high contrast lens. Such things can at least partly be corrected for in PS.

[bjanes]

The sharpening of this image was done in RSP. The use of USM in Photoshop with a large radius of 50 pixels is not a normal sharpening procedure but a technique known as 'local contrast enhancement'. Michael has a tutorial on it.

I mention this in connection with an earlier point I made, namely that a lack of contrast at low frequencies that might be more critical for subjective sharpness, is no longer such a disadvantage if one happens to have used a high resolution lens as opposed to a high contrast lens. Such things can at least partly be corrected for in PS.
[a href=\"index.php?act=findpost&pid=70899\"][{POST_SNAPBACK}][/a]

The hiraloam "sharpening" is an interesting technique, useful for minimizing haze or improving large scale light dark relationships as in Michael's example, but if you want to improve edge sharpness to compensate for a high resolution but low contrast lens, you will have to use a much smaller radius. However, the required radius for low frequence detail (say 10 lp/mm) may efface the detail at 40 lp/mm as Blatner and Fraser point out. Therefore, I think you have rested your case too soon.

[Ray]

However, the required radius for low frequence detail (say 10 lp/mm) may efface the detail at 40 lp/mm as Blatner and Fraser point out. Therefore, I think you have rested your case too soon.
[a href=\"index.php?act=findpost&pid=70902\"][{POST_SNAPBACK}][/a]

Perhaps what needs to be sorted here is my method of calculating the resolution of those feathers. If my figure of 53 lp/mm on the sensor is correct, then I do not see that the Blatner and Fraser statement is necessarily correct as it applies to this image, unless one puts special emphasis on the may efface the detail.

With this particular image, I can increase local contrast as well as dramatically improve fine detail. One obviously has to experiment for the best (or desired) effect. The first pair of crops below compare the completely unsharpend conversion in ACR with 80% sharpening at 10 pixel radius. The second pair of crops shows the original unsharpened image next to the same image sharpened twice, first at 80% and 10 pixels radius and then at 200% and 0.7 pixel radius. In the 3rd image, the already twice sharpened image has had some strong local contrast enhancement applied (30% at 50 pixel radius).

One can argue about whether or not some of these images are over-sharpened, but the point I make is that the finest detail is very much enhanced as well as the contrast.

(1) [attachment=827:attachment]    (2)  [attachment=828:attachment]   (3)  [attachment=829:attachment]


Perhaps, Bill, you can give me some examples of sharpening routines that destroy fine detail, so I can avoid them   .

[BJL]

Now before we go any further, I'm going to lay bare my mathematical reasoning. If it's flawed, I'm sure Jonathan will jump on me like a tonne of bricks and make me look like a complete twerp. BJL might be more forgiving.

The width of the 20D sensor is 15mm.
[a href=\"index.php?act=findpost&pid=70849\"][{POST_SNAPBACK}][/a]

Here is a kilo of bricks then: the image output region of the 20D sensor is 22.5mm wide (15mm is the height).


P.S. The ideas of SQF in the original post reflect the idea stated in Michael's article on reading MTF graphs, where the greatest emphasis is put on high MTF at 10lp/mm (for 35mm format), more than on the 30 or 40 lp/mm curves. This seems soewhat related to looking at what sort of image detail is reproduced with close to correct contrast level, more than looking at what is visible but at far less than natural contrast.

[Ray]

Here is a kilo of bricks then: the image output region of the 20D sensor is 22.5mm wide (15mm is the height).

Not when the image is oriented vertically, and this image was.

The ideas of SQF in the original post reflect the idea stated in Michael's article on reading MTF graphs, where the greatest emphasis is put on high MTF at 10lp/mm (for 35mm format), more than on the 30 or 40 lp/mm curves.

Nothing wrong with the idea. I'm just making the point that Photoshop has changed the significance of that. Also, as Gary mentioned, most 35mm lenses that have a good MTF response at 40 and 50 lp/mm also have a good response at 10 and 20 lp/mm. There seems to be a preferred balance. It's a lot easier to change the contrast of an image at various levels than create detail the sensor was not able to capture because of a poor MTF response of the lens at 50 lp/mm.

[bjanes]

Perhaps, Bill, you can give me some examples of sharpening routines that destroy fine detail, so I can avoid them   .
[a href=\"index.php?act=findpost&pid=70909\"][{POST_SNAPBACK}][/a]

If I had a lens with poor MTF at 10 lp/mm and good MTF at 40 lp/mm, I could test your thesis that sharpening can restore contrast at 10 lp/mm without affecting resoution at 40 lp/mm or creating undesirable sharpening artifacts. Or, you could do the same. Until that time, your thesis remains unproven despite your unsupported assertions to the contrary.

BTW, Bruce Fraser points out that you should always check sharpening at 100% with Photoshop, or else the image will be alaised. This applies to a screen view of 25% or 400%. The 400% view so favored by pixel peepers in PS is heavily alaised. Did you ever wonder why checking the resolution charts on DPReview is so frustrating if you use 400% in PS?

Here is a 400% view of such a chart in PSCS2 and also with Microsoft Picture & Fax Viewer, a simple application which does not use antialaising. It must be viewed at 100% to see the effect: