Is there an accepted technical definition of blur . . .

[xpatUSA]

. . . or, like so many photographic terms, is "blur" whatever someone thinks it is?

I was just up against a Gentleman on another site who thinks that large pixels cause blur due to the lack of resolution. Me being me, I was trying to counter with talk of edge response a la MTF slant-edge testing, or the slope of an edge profile or even a simple 100% crop. Needless to say, our discussion bore no fruit.

Meanwhile, over on slrGear, their lens tests are graphed in units of blur but finding what that means is quite a task.

While realizing that perceived blur has several causes, e.g. OOF, diffraction, mist on lens, alcoholic shakes, shooting from a speeding train, etc., - can blur be quantified with actual numbers derived from something other than opinion, for example, a mathematical equation or maybe sampling theory or summat.

What is 'blur', quantitatively speaking?

[ajohnw]

Blur all comes down to the final image. The metric optical people use is that the human eye can resolve 1/60 of a degree so it can all be worked back from that. That varies with contrast though - less contrast worsens resolution so the figure is based on 100% black and white. I've also seen mention of some fraction of the sensor diagonal but no info on final image size

Pixel size has nothing to do with it on it's own. It does in relation to sensor size. For the same field of view the focal length has to be scaled according too sensor size but the focal length also sets the "magnification" of the lens so smaller sensors need higher pixel density than larger ones to get the same resolution at a pixel level so in principle with the right focal length and the right size sensor 25mm pixels would be fine.

Diffraction is a bit strange in the photo world - in several ways. For one thing 4 pixels and interpolation are used to record colour usually RGGB. Worse still is the anti aliasing filter. It's best to google Nyquist on that subject. It's based on electronics and easier to visualise that way. If some signal varies with time say it repeats every second it has to be sample at twice that rate to reconstruct it. Believe me there is an entire rats nest behind that which assumes that the signal has a certain shape. Switching to pixels the same thing is needed. An optical filter that makes sure that the pixels don't see any light changes beyond capabilities otherwise they will alias it. In this case that means best resolution is 2 pixels and probably worse than that in practice. Interesting - 2 wide by 2 high is 4. Going back to the signal if it was sampled at 1 second intervals all readings would be the same. If sampled at 1/2 second intervals and the signal was changing at some rate faster than 1 second it would not be possible to reconstruct the signal and this is where aliasing as it's called comes into thing. If it goes wrong this pages shows what the photo's can look like in places. There was a better page showing problems without flare but I can't find it.

http://www.pages.drexel.edu/~par24/r...Artifacts.html

On top of this is the debayering algorithms. How all fit together - need to talk to some one who designs them but the chances are that a 4 pixel square or somewhat smaller than that gives a better idea of resolution. Smaller may be possible after a fashion because of the way debayering can work.

The wiki has a little on anti aliasing filter but searches for Nyquist and Nyquist frequency might give more understandable results.

http://en.wikipedia.org/wiki/Anti-aliasing_filter

John
-

[Glenn NK]

. . . or, like so many photographic terms, is "blur" whatever someone thinks it is?

What is 'blur', quantitatively speaking?

Ted:

My take on blur is that the word qualitative is more meaningful than the word quantitative.

The term measurebating comes to mind.

Glenn

[William W]

What is 'blur', quantitatively speaking?

If you take the whole picture, then minus the sharp bits, you're left with the blur.

Quantitatively:

WP - SB = B


WW

[William W]

While realizing that perceived blur has several causes, e.g. OOF, diffraction, mist on lens, alcoholic shakes, shooting from a speeding train, etc., - can blur be quantified with actual numbers derived from something other than opinion, for example, a mathematical equation or maybe sampling theory or summat.

On a serious note . . .

Yes good question. I like it!

I think (I am very sure) that older texts will indeed quantify blur associated with Subject Movement. I recall that there is chapter and verse in some of the old Kodak Manuals on 'Blur' being quantified much the same was as we quantify DoF as "acceptable sharpness" and then (those Kodak Manuals) gave the elements:

Subject Speed
Subject Distance to Camera
Subject Direction of Travel (in degrees relative to the Lens' axis)
Focal Length of Lens
Shutter Speed

and from that data we could then calculate if the (Motion) Blur would be 'noticeable' or not (on a 10 x8 full frame print held at arms distance).

I haven't seen any work which classifies ALL the causes of "blur" and then seeks to quantify each so precisely and passionately as those old Kodak Manuals described the steps for establishing the 'rules' for setting shutter speeds for 'stopping' motion.

I think that to quantify all Blur - (if we really wanted to) - we'd first need to establish a list of the Blur Types, from that point it would not be too difficult to quantify each of those, individually.

I guess when more than one blur type impacted on the final image, then that would then be an interesting final formula . . .


WW

[Black Pearl]

I suppose one could add:

Is blur always destructive?

Blur can sometimes add to an image, dramatically improve it and is therefor something to actively pursue.
How does one qualify blur in these circumstances?

[Glenn NK]

I suppose one could add:

Is blur always destructive?

Blur can sometimes add to an image, dramatically improve it and is therefor something to actively pursue.
How does one qualify blur in these circumstances?

Is blur always destructive - I don't think so. Bokeh is blur in areas that are OOF - I'm not aware that there are any measures for bokeh - it is subjective and non-quantifiable.

As noted, blur has other causes - perhaps this indicates that measuring blur will be a bit difficult.

How does one QUANTIFY blur? Good question, and rather challenging I think.

If someone cares to undertake developing a blur measurement system, I would keenly follow their progress. This could be construed as a friendly dare.

[xpatUSA]

How does one QUANTIFY blur? Good question, and rather challenging I think.

If someone cares to undertake developing a blur measurement system, I would keenly follow their progress. This could be construed as a friendly dare.

Well, I found a site where it has been done! Even accounts for a few types of blur. That would be the good news.

The bad news, it has stuff like this:



Too much, even for a non Real-Worlder like myself, see here:

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

The early part of the article on sharpness is most instructive but the section on blur prompts me to withdraw the question in the OP. In other words, having found somewhere where blur is actually quantified, I now wish that I hadn't and may yet "get a life" and "go out there" into the Real World and shoot some pictures

[Richard Lundberg]

Excellent article. You could probably say the blur is the width of the degraded point spread function in lines of ideal resolution, or the number of lines of ideal resolution that 3 sigma of the degraded point spread function occupies, or the per cent reduction of the MTF50.

[FootLoose]

Or more simply, the bits that I can't see clearly

[ajohnw]

Not bad but it looks like he assumes diffraction limited optics to me. Good to see 2p crop up when colour is introduced though and I did wonder how they actually anti aliased.

As it's of great interest to some aspects of photo's mainly through a microscope I do have a couple of graphs about.

This one is plots of MTF curves with a bit of explanation. Curve a can be seen as a perfect lens. All perfect optics have the same shape curve. All that changes is the scale at the bottom. The vertical scale is contrast out for 100% contrast in. Where the plot reaches 0 contrast is set by the F ratio of the optics. The shape of the curve is always the same.



Curve C is an oddity but believe it or not mirror lenses can have the higher resolution curve shown. Doubt if the ones we buy can though,

Curve a in b is where realistic camera lenses lie in the scheme of things for the reasons given. That's down to the fact that there are lots of lenses in them and the accuracy that needed in perfect optics. The accuracy needed is illustrated by this one.



Curve b is what Rayliegh's Criterion represents. Basically errors in the refractive index, distances between lenses and the glass thicknesses in total mustn't exceed 1/4 wavelength of light , about 0.14 um in green light. There are some very influential pundits about though that wonder about curves e and f and a bit about d due to actual measurements. It seems a,d and c are ok. Personally I feel that the whole thing stinks and only indicates trends as when lenses are designed the density of light across the circle of confusion can vary over wide limits. It might be very even or more intense in the centre. The shape can vary too and then there is colour variations in both. This also makes a bit of a mockery of the test described in the link but it's better than nothing.

John
-

[ajohnw]

Perhaps the easiest thing to do is to follow Canon and others. Their DOF tables seem to assume a circle of confusion of 0.035mm on the sensor for full frame expanded to fill 8in of the usual 10x8. The human eye resolution way would come up with a figure of under 1/2 that. Figures in the range of 0.030mm have been used and even as low as 0.024mm but the latter it seems only in Europe. The larger number works out at about 1/1400 of the sensor diagonal.

These decisions were made on the basis of aberrations in lenses and "problems" with film. In other words in the real world it wasn't possible to meet the human eye criteria. I don't entirely believe the film aspect. 35mm was greeted with derision initially because everyone used something a lot bigger. It had it's problems so the usual poetic license might have been used.

Canon still have a bit of info on it. Not aware of any actual DOF tables but if about a web dof calculator could be used to determine what they assume.

http://cpn.canon-europe.com/content/...th_of_field.do

John
-.

[Inkanyezi]

Circles of confusion cause a lot of confusion, but it is rather clear that it is a valid consideration sometimes.

But in practise, constraints within the real world sometimes are more influential than the circle of confusion. When the spatial frequency of the target is too large, and this is far before we hit the Nyqvist theorem, blur occurs anyway, so that objects with lower spatial frequency nearby may appear sharp, while those with high spatial frequency become blurred. The image is in the original 16 megapixel size from the camera, so if opened separately it can be pixel peeped. It is however saved with much compression due to constraints in storage space where it is posted.

I don't see the objective for quantifying this blur. It is the blur that makes nearby objects in a close-up image of a landscape appear sharper than foliage and wave ripple farther away, even though actual focus may be in the lake at the middle ground and the DOF table clearly indicates that DOF is from the stones at the shoreline one metre away to infinity. The foliage of vegetation on the other side of the pond has too high spatial frequency, and it will not appear as sharp as the closer objects, stones at the shoreline, even though their circle of confusion may be larger.

In the image I have posted here, there are several different manifestations of blur, the only absent one motion blur. There is colour moiré, as well as blur from uneven focal plane of the lens, and there are places where spatial frequency is too high, as well as very small depth of field close to the camera. So focus is set approximately along the floor, but the faraway millimetre marks beyond 130 cm from the camera cannot be discerned, and the closest part of the tape rule is blurred because of the very shallow depth of field. However at that close distance, spatial frequency is much lower, and millimetre marks can be clearly discerned.

The zero end of the tape measure lies directly under the camera.
Taken with Schneider Xenar 50 mm f/2.8 mounted with Hartblei tilt adapter on OM-D E-M5 and tilted 5.5 degrees forward.

And maybe I ought to take a turn with the vacuum cleaner.

[ajohnw]

That's an interesting image Urban. I'd guess some of the mm marking loss is down to the angle, software, pixels etc but I reckon I can see a slow loss of contrast in the fine inch markings. MTF type effect. Had to download it and swing it through 90 degrees to see this.

It just goes to show that in some ways the detail in the subject comes into it and probably software.as well. The carpet looks fine to me.

John
-

[xpatUSA]

Not bad but it looks like he assumes diffraction limited optics to me. Good to see 2p crop up when colour is introduced though and I did wonder how they actually anti aliased.

As it's of great interest to some aspects of photo's mainly through a microscope I do have a couple of graphs about.
-

Here's my contribution to graphery - I've posted it before but it fits here, I reckon.



Check out the difference for f/5.6 between top-center and top-right!

With out new-found knowledge, we should be able to pick the best of those lens for the job in hand, eh?

[ajohnw]

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

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

John
-

[Brev00]

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.

[ajohnw]

The easiest thing to do is to completely ignore any maths, check in several places and then make your own mind up with your camera.

John
-

[Inkanyezi]

That's an interesting image Urban. I'd guess some of the mm marking loss is down to the angle, software, pixels etc but I reckon I can see a slow loss of contrast in the fine inch markings. MTF type effect. ...

Tilting is unfavourable toward the edge of the sensor in the direction one tilts. MTF surely decreases, and this is a Tessar-type lens made around 1960. I am surprised it has good enough coating, but I guess it is also a matter of very few elements. Many lenses from the sixties cause contrast loss in the centre of the image due to reflections from the AA filter bouncing back from the rear.

But subject detail often is more important than circle of confusion regarding perceived sharpness. The horizon in a wide angle shot can never appear as sharp as the close foreground, even though it might be perfectly in focus. Setting hyperfocal distance may be counter-productive to sharpness, as very distant objects won't appear sharp anyway, and it sacrifices foreground sharpness. It might often be better to use a DOF from say 0.5 m to 10 m for a wide angle lens, than to aim for infinity. Grass, foliage and ripple wouldn't become sharp more distant than that anyway, and overall sharpness can be perceived as crisper, as with the closer foreground more in focus, more of the image will be sharp. This is particularly true for very small sensors and short focal length. A full frame camera would not suffer as much as the µ4/3 from this effect, as it would use a 100 mm lens for the same angle as I used 50 mm. Even better if one could have a medium frame camera. I have seen what a Rollei SL66 can do with 80 mm Planar and 150 mm Sonnar with tilt. It blows the puny 4/3 format away in this respect. However, even though I could get the camera, I couldn't afford the back.

[ajohnw]

The reflection of even 60's glass could well be down round the 1% level and due to make they may have been using multi coating even then. What has happened in this area is interesting. Lens manufacturers would often only coat the surfaces that were visible. More exotic glasses have put and end to that and maybe guilt too. Some early optical glasses can suffer detrimental effects from moisture as well.

On the shot the contrast drop off could be mostly down to the off axis performance of the lens. Such lenses usually reach their best across the field at F5.6 / F8.

Ted's graphs are plotted in a way that seems to be favourite in the camera manufacturers world. They measure mtf at say 3 specified line pairs per mm. By measure mtf I assume they mean measure contrast levels and then plot % of theoretical max but it's common to assume that the Rayleight line is theoretical max - diffraction limited which it isn't.. The other axis is across the lens. That way they can show the results vertically and horizontally across the lens as these are very likely to differ. They probably pay a lot more attention to that on some lenses which are intended for very specific uses. Possibly lenses like this one - I haven't compared just guessing

https://www.schneideroptics.com/ecom...=1822&IID=8296

They have based their plots around 15,30 and 60 lp/mm. Could be others don't go that fine.

John
-