Is there an accepted technical definition of blur . . .

[Inkanyezi]

My friend with the Rollei had an APO Ronar for her Sinar P, but she sold it with the camera and is not really crying over it, as it was too long for the smallish medium format back anyway (240 mm). She says she'll get an APO Sironar. The Rollei has focusing in the camera body as well as tilt, so it will only need a very simple adapter.

I also was thinking along those same lines, but I haven't found a good focusing mechanism to adapt very short lenses, and the EL Nikkor had too bad AR coating, so it was useless, just as the Agfa Repromaster lenses that I once had and tried. My best lens, regarding contrast and resolution is a Trinar 50 mm, and it also has a rather flat field, but there is no focusing gear. So I keep looking for lenses to adapt, and I just recently found the Schneider Xenar. It is mistreated, but optically sound and the focusing helix is smooth.

[Inkanyezi]

I always considered blur the last refuge for non-quantifying folk like myself. My incredible ability to skip graphs and charts has temporarily kept it that way for me. My inability to understand it will provide future proofing.

I think that is a sound stance on the phenomenon. To me, it is something to investigate, finding out the properties of the tools, so I can use them with confidence, knowing what the result may be. I think that it is in a way as how we learn to swim. The only reasonable way to do it is in the water. One must understand the blur and what causes it, and how as well as where, it can be avoided. It is also rational to find out where it cannot be avoided (or shouldn't) and take that too in consideration when making the picture.

But quantifying?

[xpatUSA]

They are unusual graphs with mtf on the vertical axis. What book is it out of? I don't believe it.

Unusual? Did you not post this one?



The x-axis 'S' of the graphs I posted is normalized frequency v/Vo, same as yours above.

Try this one. None of them are anywhere near the Rayleigh spec. The lens must be cheap tat.

http://www.schneideroptics.com/pdfs/...xl_56_47_2.pdf
-

The link represents a classic PO move (photographic obfuscation) showing MTF versus picture height whereas I posted MTF versus normalized frequency.

I posted the graphs for information. I have no idea what book they came out of but the authors are well-respected and credible. Might have been "ROSENHAUER, K., and ROSENBRUCH, E., 1967, Rep. Prog. Phys., 30, 1."

If someone chooses not to believe such august gentlemen, that would be his privilege

cheers,

[Inkanyezi]

I think that "blur" in a way is quantified as MTF, but the impact in real life, or rather on an actual image, is difficult to fathom.

Mostly when sharpness is discussed, "in focus" would be apprehended as sharp. However in the actual image, too high spatial frequency may render its detail indiscernible. Thus, quantifying sharpness objectively in a way that relates to the image makes sense.

I think that the problem here is not that we don't have a quantifier, but that it is difficult to understand how spatial frequency varies, and how its variation within an image influences sharpness. MTF can be applied to other types of blur than optical defects, although the actual photographic situation is not simulated in the laboratory, but the curves are made in an "ideal" setup. Just as defocusing may alter the MTF, and is sometimes measured, also motion, vibration and angular movement around various axis affect MTF of the system in real life photography. In simple terms, it is difficult to understand the impact in the actual image, even though it can be quantified.

[ajohnw]

The Schneider one looks like a true mtf to me Ted.- after a fashion as the way it's presented it's derived as I mentioned and shows radial and tangential plots and even gives the spatial frequencies used. This site uses the picture height method as most do but this one has numbers.

http://www.photozone.de/all-tests

Zeiss have some information available on how the Schneider types are calculated - also show a fabulous optic. Oh yeh. Schneider don't seem to always offer them on 35mm. I wonder why.

As to blur these days I don't think it's possible to truly define it especially critical levels such as hyperfocal aspects but the basic ideas still hold. The relationship between the blur on the sensor and it's size when viewed. Most cameras interpolate the image before we can even look at it. PP is often sharpen in and sharpen out - after the shot has been reduced. Contrast etc in one way or the other is also messed with.

Some cameras live in Pixel Density Fairy Land. One aspect of that is that lenses themselves could well be acting as anti aliasing filters so there is no need to put one on the sensor. I'm sure my EM-5 lives in PDFL with the 12-50mm on. I should really try the 14-42mm on it as that definitely wasn't at 12mp. Compacts have been in PDFL for years despite the fact that their lenses are tiny which really can result in better optics at a lower cost. The ultimate small optic is a microscope objective in that respect. Sometimes they are nearly diffraction limited but the contrast level is so low at that resolution slides are stained to boost it. Good 'ens cost a fortunes as some aspects have to be done by hand even at that size.

John
-

[Inkanyezi]

Just to make it clear, we can use MTF defined in different ways. Mostly it is measured with some measuring apparatus that sets the limits, but it can also be measured as the MTF of the entire system, i.e. it is the finished image, printed or on screen, that defines it. Then many other factors than the optical and sensor-based ones come into play, so it should be clear just what MTF we are relating to. Measured directly in optical measuring apparatus, measured in the electrical levels of sensels on a sensor, or measured in an image from that sensor after data treatment. In real life, it's the impact on the image that is of most interest.

No matter which way we see it, lens tilting always puts some part of the image sensor in a less favourable area regarding the MTF of the lens, while on the other hand, another part of the image will be more favoured. It is the area around the optical axis that has highest MTF, and the farther away from that, more degradation. This spot moves when we tilt, and it moves toward the closer objects in the plane of focus.

[ajohnw]

MTF is a design tool. It's calculated from wave front errors and gives a very quick over view of performance and as such it belongs in optical design software tools really. Where it's used for testing results are extrapolated from a single or several measurement. The idea comes from electrical engineering. A Fourier series. Best explained here

http://en.wikipedia.org/wiki/Fourier_series

As more and more harmonic are added the rise time gets faster and faster. There is also an amazing mathematical idea used to analyse electronic circuits. A Dirac pulse. At one instance it's 0 and in the same instance it not. Usually referred to as having an infinite rise time and as such must contain all suitable wave forms - right out to something which is infinitely fast.

When it comes to using this for testing lenses there are 2 choices. One is a series of black to white lines that pass light in an exact sinusoidal manner or a step edge that tries to behave like a Dirac pulse. An abrupt transition from all light to no light. In one case the change in contrast is measured and in the 2nd the rise time of the light from all light to no light is measured. It's a bit difficult to do either in practice.

Pixel densities will have an "MTF" after a fashion. A straight line until the spatial frequency gets too high for the pixel pitch and then they will alias. Covered here

http://en.wikipedia.org/wiki/Aliasing

As pixel density gets higher and higher eventually the lens on the camera will behave as an anti aliasing filter that can probably be augmented by some form of software filtering. The ideal would be contrast reaches zero as shown by an MTF before the sampling frequency of the pixels is reached. - could be that PS will have to come up with new raw development software at some point. There have been cases where it does fall over but the camera jpg is ok. Unlikely to be met in practice as it's associated with weird patterns but ............ There could also be a case where the "spot" size of the lens covers all 4 of the colour pixels which means that the image spreading currently used to hopefully evenly illuminate them all isn't needed any more.

John
-

[xpatUSA]

The Schneider one looks like a true mtf to me Ted.
-

I didn't say it wasn't, John.

So, are graphs with MTF or "modulation [%]" in the Y-axis still unusual, or are you now a true believer?

TTFN,

[Inkanyezi]

On the levels where it influences the image, the pixel density graph will not be a straight line, as there are (in real life) two factors that cannot be ignored, both of them called bleeding, although technically they may have different causes. There is optical, as well as electrical reasons for bleeding, and when we come down to sensel level, it is not negligible.

Therefore, a certain spread of the "point" will occur, no matter how hard we try to isolate it from the one nearby. This is not really bad, as a higher pixel density might more truly render the MTF of the lens in electical signal levels.

But when we look at a finished image, other factors also influence appearance of sharpness, and sharpness cannot be defined equally at the pixel level as when we look at the image from a reasonable viewing distance. There are artefacts in my image from mainly two software factors, one of them jpeg compression, which is heavy, the other sharpening, which has been applied.

Sharpening, as we know it in software, is intended to exaggerate edge recognition by increasing the difference in brightness and/or colour at edges of elements in the image. It causes artefacts, that are visible at some distance from the edge. On the other hand, without sharpening, the image seems less sharp, even though its resolution mostly is somewhat higher than after sharpening.

Sharpening has no effect on areas where spatial frequency is too high for recognition of separate elements.

Some of the errors are also "sharpened" with the instrument, causing for example colour moiré artefacts to appear at a distance from the image element that has colour fringing, and that effect is discernible at several points in my image. It is not clear to how large extent colour fringing would depend on each one of the three different factors that may cause them: optical correction errors, angle of incidence toward the sensor or demosaicing.

In real life photography, it is the end result that is of most interest to the photographer, not the theory, although theory may help in understanding the problems and finding solutions or workarounds, or simply accepting that nothing is perfect. For the artist, it is part of the process to learn how to use the tools to reach intended purposes. Photography is a visual art, and for the photographic artist, it is how technique relates to vision that is of paramount interest. For many of us, delving too deeply in technostuff is not necessary, but we can just see, that some roads lead to the intended goal, some do not. We must understand that spatial frequency of the image elements and their size in the image are factors that influence perception of the image, and adapt to that knowledge. Knowing that image elements far away with high spatial frequency can never be rendered "sharp" in one of the meanings of sharp, is soothing, knowing that we need not hunt for what cannot be achieved. Then we are free to use the effect creatively.

[xpatUSA]

People say that a little knowledge is a dangerous thing. The corollary being that a lot of knowledge is a good thing. A gentleman over on LL knows an awful lot about blur and sharpening and stuff. Some of his published stuff hurts my brain. For example, this about capture sharpening. However, it does let him get from this blurred image:



To this:



Which, for initial "capture" sharpening, is pretty good I reckon. The sharpening radius was not guessed or arrived at empirically; it was determined exactly by a camera/lens slant-edge test.

Not necessary for "Real Life" holiday snaps, of course, but if if you're competing with others for a product or nature photography contract you could gain an "edge" using an edge

[ajohnw]

I didn't say it wasn't, John.

So, are graphs with MTF or "modulation [%]" in the Y-axis still unusual, or are you now a true believer?

TTFN,

No Ted as far as the graphs showing perfection at F11 I'm not. Optics have been a very very long term hobby of mine and I do own some very nearly diffraction limited optics. People photograph all sorts of things via camera lenses as well.

I seems Pentax have found an anti alias method that doesn't use filter. The move the sensor about. Seems means I have read. Interesting idea though.

John
-

[Inkanyezi]

I seems Pentax have found an anti alias method that doesn't use filter. The move the sensor about. Seems means I have read. Interesting idea though.

And the camera is already sold to the public, K-3. And like other Pentax cameras, it is priceworthy compared to the competition.

The sensor is 24 megapixels and it uses the same system that moves the sensor for stabilisation and dust-cleaning to achieve the AA effect. Ricoh seems to be a good companion to Pentax, and I expect we may soon see more cameras with similar solutions. There are a few manufacturers that could do it already with just a firmware update, mainly Sony and Olympus. The AA sensor shake may be enabled/disabled by the user. So we have a 24 megapixel APS-C without AA filter, where when the photographer wants AA filtering it can be activated. BTW it is a similar solution that Hasselblad uses to double resolution in their back, so that is also a possibility when using sensor shift. It can be implemented in Olympus, Pentax and Sony just by removing the AA filter and adding functions to the firmware.

[xpatUSA]

No Ted as far as the graphs showing perfection at F11 I'm not [a true believer]. Optics have been a very very long term hobby of mine and I do own some very nearly diffraction limited optics.

It seems there may be a misunderstanding of the graphs that I posted, if those are indeed the graphs to which you refer. I'll try to explain the X-axis where it says 'S'. S is normalized. It represents a frequency ratio, where 1 = the zero MTF or cut-off frequency at any f-number graphed. Otherwise, there would have to be a different graph for each f-number. As the f-number increases, it's curve should and does approach the ideal for that aperture setting. What you missed was that, for a high f-number, the actual cut-off frequency is quite low.

For f/11, the cut-off frequency Vo is 167 lp/mm, which is not particularly high compared to say f/2.8 where Vo is 655 lp/mm (perfect lens, of course). The graphs I posted do tell the correct story: for high f-numbers, lenses approach the ideal; for low f-numbers, real lenses are far from perfect.

Image repeated, lest we forgot . . .

People photograph all sorts of things via camera lenses as well.

Good one, John - I do miss good old English sarcasm, people don't appreciate it much over here

[Inkanyezi]

I'll take the opportunity also to flick in here, that "diffraction limited" is an optical engineering term that is also used rather loosely in the photographic community in a slightly different way. We don't use diffraction limited optics, but as John says, we photograph all sorts of things via camera lenses.

But there is an instance where diffraction comes into play, and it is when we stop down, as the diaphragm closes to a very small orifice, where the light has to pass. Diffraction does influence the sharpness of images at f-stops that are smaller than 5.6, and it can be severe at very small apertures, as f/22 and smaller, making the lens more like a hole camera and softening the image.

The simple calculation of refraction however is not true for its occurrence in real optics, as their construction may permit a larger or smaller aperture size than the virtual one. In a retrofocus wide angle lens, the real size of the aperture is larger than its focal length divided by the f-number, and in a telephoto lens the opposite. So f/32 in a 16 mm lens is not a half millimetre, but can be as large as 2 millimetres, and thus diffraction products softening the image are less than might be expected. The half millimetre aperture is virtual, it is what you see when you look at the lens from front, its entrance pupil. Inside the optical system, its physical size is larger. The wavelengths of light however are the same when they pass that orifice, and they don't diffract as much through a 2 millimetres hole as through a ½ mm hole. Signal, light propagating in a straight line as in the ray theory, thus is more favourable compared to noise (diffraction).

Also, the diffraction that is relevant, as we use camera lenses, is nowhere near the diffraction limit, but it is the relative amount of diffraction products from the aperture compared to the light rays that pass straight. Diffraction can decrease contrast significantly when using very small apertures, and softening is worse where spatial frequency is high, so that also resolution/sharpness decreases visibly. This is what is measured in MTF testing of photographic lenses, defining their "sweet spot".

The simplistic way of coping with it is to accept that it can be measured at f/8 and becomes worse at smaller apertures for all lenses. So stopping down more than f/8 does not increase sharpness where focus is set, even though depth of field increases.

[ajohnw]

Some would dispute that diffraction comes into play Urban. To me that indicates the likely size of circles of confusion. It fits in with my own "normal" photography experiences as well. Here's a web page that goes through some of the problems with data and reckons it does come into things.

http://www.kenrockwell.com/tech/mtf.htm

I feel the need for a test where I can be sure every other reason is ruled out. An EM-5 is ideal for that as a magnified view can be used that truly shows what the pixels are seeing. That rules out focusing aspects - I hope. There doesn't seem to be a magnification that causes pixels on the sensor to match those in the view finder. However that may be better than either AF or using the viewfinder.

The other and probably more meaningful way of testing a lens is to google USAF resolution targets. It may still be possible to download and print them. Last time I was interested in them I didn't notice any for sale. There are more usage instructions about. Basically the distance from target to lens is set to achieve the lp/mm range needed. Some magazines used to give test targets away for free. Many concluded that lenses sent for reviews were selected. Personally I think lenses vary and when tested to the limit differences will show. Focusing accurately at diffraction limit levels still wont be easy.


I'd allready mentioned that all diffraction limited mtf curves have the same shape Ted. 166 lp/mm? - why do Schneider use much lower figures and why do the graphs start low and tend to remain at the same level. The reason goes back to the graph I posted that shows a poorer MTF that maintains the same contrast for longer - good photo lens practice. There are a couple of m 4/3 lenses kicking about one of which tends to show that too much even contrast isn't a good idea but a lot depends on the size of the final image. If the size is such that the detail is lost it doesn't matter to some.

From various aspects seen in other areas I feel that the slant edge test tends to get unreliable as things move away from diffraction limits. In ray tracing term diffraction limit means all rays disappear into a zero sized spot within the usual diffraction disc This implies that the optical path length is the same for all rays as well. That's distance measured in light waves. As things move away from that the the mtf deteriorate more rapidly with increasing errors - that's what the defocus mtf's I posted show. In practice it's hard to meet the 1/4 wave one. Obscuration in some areas makes Shrehl ratios favourite and no mention of diffraction. Photographically that along with a certain sized circle of confusion makes a lot of sense. Visually - telescopes - seasoned observers can detect errors of less than a 1/10 wave.

John
-

[Inkanyezi]

Verbosity to me boils down to diffraction being an important factor in astronomical telescopes and in microscopes, while for photography we can use empirically found out "sharpness" being the practical consideration. Measurbation is not to much avail, other than it can indicate that there is a difference between different optics, particularly those figures off centre where MTF decreases. We need not know what causes it, but it's certainly not diffraction as long as the aperture is f/5.6 or wider.

Nevertheless, MTF of the whole system, from the capture of an image, down to the finished print, is one measure of "sharpness", for anyone that wants one. Whether it is meaningful is another matter; spatial frequency of subject matter and its distance in relation to the focal length are also determinants, but I don't think a number would be meaningful. Our visual assessment is quite satisfactory for all artistic needs.

[xpatUSA]

An excellent explanation from Urban in his precious post, leaves little more to be said!

167 lp/mm? [for f/11]

Zero MTF frequency = 1/(wavelength * f-number)

where f-number = real focal length/actual aperture diameter (thanks, Urban)

1 / (545e-9 * 11) = 166806 lp/m = 167 lp/mm

(545nm was in my spreadsheet when I opened it up, should perhaps be 555nm often used for theoretical 'green').



Now, Falk Lumo says:

The Rayleigh criterion defines the limiting resolution by a line pattern where two line pairs are r_Airy apart. The true limit though is
f₀= 1 / (λN)
which is the frequency of zero MTF. The full MTF is:

MTF_diff (f) = 2 / π(arccos(k) - k √(1 – k²)) with k = f/f₀= f λN

Ref: http://www.falklumo.com/lumolabs/articles/sharpness/

If you find Lumo a little hard to follow, creating a tendency to disbelieve, Bob Atkins' stuff is presented more clearly:



Lumo's formula gives 450 lp/mm same as in Atkins' diagram above. Looking at the diagram, with its easy-to-understand lp/mm axis, can you guess where a "Perfect" f11 lens might fall?

Ref: http://www.bobatkins.com/photography.../mtf/mtf2.html - where you will see two other familiar diagrams

Waiting for your proof that 167 lp/mm is an incorrect zero MTF frequency for light at 545nm and a perfect lens setting of f/11 . . . . . . . . .

[ajohnw]

It's worth throwing one more aspect in. Like diffraction it's factual but throw in interpolation and debayerig things get a lot less clear.

The diffraction spot radius is 1.22 x wavelength x F ratio. Different wavelengths are sometimes used which can be a bit of a pain but lets say an F6 set up has one of 8.2um diameter. That's green light, red is bigger, blue smaller. Any F6 optical system can produce diffraction spots of that size if sufficiently accurately made. Along side that comes resolution. The angular resolution of an optical system is 1.22 x wavelength / diameter. To keep that simple the result is in radians but that doesn't really matter. These are fundamental. The focal ratio is focal length / diameter. So if a system has a crop factor of 2 compared with another and both are F6 or any other value with the same angle of view the cropped one can only have 1/2 of the resolution of the other one. The image is smaller so there is less room for diffraction spots. It's generally accepted that images are made up of myriads of diffractions spots otherwise optics would fall over and bang it's head.

This doesn't mean throw away crop factors of 2 or even 1.6 as pixels come into it as well. And the practicalities of producing multi element lenses in quantity. Small helps within limits. Sticking with a crop factor of 2 the larger format lens could be simply scaled up by a factor of 2. It would still be F6 but unfortunately the light rays have to travel further so the tolerance of the bits that bend the light to get it to focus has to be tighter. This often means more pieces of glass in some applications. Trouble with that is that tolerances stack up as well.

Resolution, blur, sharpness are all the same things out of PP packages and basically it's all swings and roundabouts and what the designer aims to do and what the production facilities can manage. Cost bears heavily on the latter even down to the type and quality of glass used.

Not much chance of measuring or calculating the equivalent diameter of a camera lens as most have some vignetting. That can be used to improve blur across the field. More swings and roundabouts. It has other effects as well when the lens is wide open. Yet more ...............

John
-

[Inkanyezi]

I love those freudian slips. :-)

[ajohnw]

I'm not arguing Ted other than maybe the practicalities of perfect camera lenses even at F11. One day I will figure out a way of measuring their performance accurately but I'm not going to stick resolution targets on the fence at the bottom of the garden again as I am more interested in how they perform when I take a photo and to me diffraction effects don't really figure not spot sized ones anyway. However I feel the need for a test coming on. It's more feasible with mirrorless.

Why I can go on about this subject sometimes is down to a long term hobby. When feeling really manic if that's the right word I design optics. I've come up with one amazing one. Problem part aimed at full frame and looks like I'm going to stick to m 4/3. Another problem - is it possible to keep the parts it uses in place with sufficient accuracy. Another design involved totally disregarding convention in just how a particular type of optic is optimised. Lots of failures as well especially on apo refractors but then I couldn't afford the glass these tend to use anyway. It's generally cheaper to buy a telescope that uses them. APO is also a much missused word. There are apo's and APO's

After mentioning full frame I should add that an around 3,500mm focal length F16 arrangement isn't everybody's cup of tea except maybe the paparazzi but it;s a pretty extreme achievement at home. I did think of posting it but it wouldn't add much to this thread.

John
-