« Normal Lens = Universal Lens? | Main | New in the Archives: The Case Against Zooms »

Saturday, 28 July 2007


Feed You can follow this conversation by subscribing to the comment feed for this post.

I don't want to play this down, since I think Fuji is doing an excellent job with these cameras. But while Mike and Thom have primarily emphasized the sensor size, their other major concern is responsiveness/speed. Part of what makes these cameras appealing is that you can take them with you anywhere, ready to take a shot when the perfect moment presents itself. If the camera has significant shutter lag, however, the sensor won't help you get back that missed shot, no matter how big it is...

Anyway, I'm keeping my fingers crossed and looking forward to the reviews. In the mean time, my Olympus XA is serving me well...

No RAW, sigh.

The lens is 35-105mm in 35mm equivalent. That's just not wide enough for me, and for many others I imagine. Still it's good to keep an eye on what Fuji is doing with their small cameras. The small cameras I know of with IS and wider angle lenses could definitely benefit from better low light performance and better dynamic range in general.

What I've seen so far in full size image samples from several digicams in several pixel counts, is simple: small sensors get a lot less fine detail, so yes, you get a huge file, filled with mushy tree branches or people hairs. There's simple no comparison. So, megapixels are a surface measure unit, not a resolution one.

The truly bizzare thing about the lens choice is that Fuji is entirely capable of putting a 28mm equivalent lens in their compacts, not just their superzooms as the concurrently announced F480 demonstrates.

As far as noise goes, with 12MP, I'm not expecting miracles. The headlining high-ISO specs are all pixel binned results. (ISO3200 @ 6MP, ISO6400 @ 3MP) Based on it appears that Fuji is tacitly conceding that the noise output from a small sensor with 12MP is too much. It's a shame because one can't help feeling that had they stuck with 6MP, the results had the potential to be nothing short of revolutionary.

The addition of A- and S- priority modes is a welcome addition, although contrary to the post above, their official website makes no mention of a true "manual" exposure mode. I'm very curious to see in the reviews if they've added histograms (at least review) to the interface.

I've watched this series since the original F10. Watching the improvements they've made to the sensor over this short period of time is truly astounding. That being said, I've only ever watched because the Achilles' heel of this series has always been the actual camera side of things. Improvements have only ever come in baffling fits and starts, without very much rhyme or reason. Which is all the more damning because, for those who remember, the now defunct E-series of Fuji compacts demonstrated quite nicely that they are certainly capable of designing a nice "camera" without the glaring short-comings of the F-series models.

Unfortunately, the claim of "full manual control" is probably not correct. The modes indicate shutter-priority and aperture-priority. If the "M" mode is like the Fuji F31, it does not let you set the shutter speed/aperture/ISO all independently, as you would expect. It is a complete misnomer of a mode setting. You essentially have to use exposure compensation to arrive at a correct exposure in tricky lighting situations. If only there was a separate autoexposure lock button it would take care of the majority of these cases, but there is not... You can read more about the F31 limitations (and desirables) at my review:


I would be delighted to discover that the F50 actually offers a true manual mode.

What very few compacts have is a specific design bias to the HAND -- can you learn to operate the camera entirely by touch for everything but the aiming? This is a huge part of the experience of using mechanical compacts. Any time a choice is made from a menu or an LCD display, the hand is slighted, the use is slowed, the value of skill diminished. People who lament the lack of a DMD (well, me) want a camera that lets them make decisions with their hands without distracting their eye from the subject.

Eric brings up a good point regarding how "fully manual" the "full manual control" actually is. A quick re-read of Fuji's product announcement, as well as a search through a number of sites echoing the "full manual photographic control" claim does not clarify matters. Aperture and shutter priority modes are close, but not the same thing. Unfortunately, Fuji Japan is on Summer Holiday from July 28 through August 5. Fuji USA might have the real scoop, I'll check that out...

No RAW, at least twice the linear resolution a camera that size should have, and no possibility of an even remotely clean ISO800 (when I can accept noise reduction artifacts, I'm quite capable of applying them myself in postprocessing).

But the main premise is faulty. Shake reduction is not an alternative to high ISO. It is not an alternative to large apertures. It is a complement to both. All three, separately, are important for low-light photography. It's like steak, gravy and potatoes: all three need to be well-made for a great dinner. What you're saying here is that the gravy is extra good on this camera, so it doesn't matter that the steak is kind of tough and the potatoes are rubbery.

It's a neat addition (one that competing makers have had for some time, of course), but it's not some miracle, and it is definitely not a replacement for a sensor with actual low-light capability.

At least at ISO 100 as they are, I think those image samples are really impressive.

I love my F10. If performance at ISO 400, and of the IS is good, I might get this one.

I don't miss viewfinders. There was never a good one on a compact camera anyway. I like using an LCD. More flexibility of viewpoint, even if it's not tiltable (which is best).

Janne has repeated the concerns regarding the pixel packing that are already mentioned in the post. It is premature, however, to declare that "no possibility of an even remotely clean" ISO 800 capture exists. Current Fuji compacts sporting the Super CCD have remarkably clean capture at ISO 400, and unexpectedly solid performance at ISO 800. Let's wait for samples.

This camera is unique among compacts in combining both shake reduction and (based upon lineage) class-leading high ISO capabilities. This is not a miracle, but it is new. Compare Fuji F20/30/40 results at ISO 400, 800 and 1600 with those of the "serious"Canon G7 (which has a Sony 10 MP sensor). Most eyes quickly concede the edge to Fuji. Now add sensor-shift VR for the first time with the Fuji sensor...that is where the F50fd generates keen anticipation.

As for the third leg of Janne's triple crown, large aperature, most dSLR's are shipped with a kit lens starting at f/3.5, correct? The supremely pocketable Fuji provides a brighter f/2.8 that has already demonstated great corner-to-corner sharpness. Let's wait until dinner is served before we pan it.

Should you wish that you had the large sensor Sigma DP-1 with the 28 prime f/4 & RAW, or the zoom lens from tiny sensor Fuji w/ face detector, if Elvis suddenly walked in front of you? (or do you think that the 50 Summilux/Leica M is the only combo that could handle it?)
Seriously, many photographers prefer compact cameras for regular work. And, as most people posting here agree on, the perfect compact size camera does not exist. In a couple of years it may come. Or it may not come. But perhaps some photographers could get everything they would ever want with TWO compact cameras?
Think about it: The point of having an SLR, is – or so they say – to be able to connect a wide range of lenses to the body, from extreme wide angle to monster tele, including zooms and prime lenses, big & fast and compact & slow, macro & tilt/shift. And the point of buying a point & shoot is – according to the same convention – to have an all-in-one small camera.
Why is it then that a lot of photographers commenting on TOP tell us that they use their SLRs or DSLR with only ONE prime lens? Because it fits their needs. Some of them even tell us that they use the big and expensive Canon 5D with one lens, a 50/1.4 or a 35/2 or 1.4. Because they like the size and weight of that camera, the big sensor, the picture quality, the big view finder, or whatever. In any case, they brake the rules.
So, there is nothing against thinking in the opposite direction: getting two or three compact cameras as a kind of “system”, much like the way we think of SLRs as system cameras. (Not only because of size, but perhaps also because of the way the small sensors draw the image, and the large depth of field. Some people prefer that.) Accordingly, if you need tele, but also wide angle, you could get a compact wide angle camera like, say the Ricoh GX100 with the 24-70 mm lens AND a Canon G7 (35-210 mm) or an even more extreme zoom lens (10X, 18X, whatever). Think of it as a very compact camera system (and not a system camera).
Speaking of the Fuji 50, here is my suggestion for a potentially superb “camera system”, for photographers who enjoy small cameras (and can afford to buy two for, say, the same prise as one entry level DSLR and a couple of small lenses).

1) The not-yet-released Sigma DP-1 for street use and landscape (28mm wide angle, more manual control, RAW and better dynamic range (because of the bigger APS-sensor)).
2) The Fuji 50 (or 20/30/40) for low light and zoom range (35-108mm).

Some times you just grab one of the cameras for a few hours of shooting (or you keep it permanently in your pocket just in case Elvis, or Jesus, shows up somewhere). But when you intend to do outdoor and indoor shooting, you may bring both of them. Not worse then carrying a rangefinder and one extra lens, is it? And much more compact and unobtrusive than any DSLR with a zoom and a prime.
Since the only compact cameras with prime lenses are the Ricoh GR-D (with the 28 mm lens, and an optional, nice 21 mm extention), and the soon to come Sigma DP1, the most interesting thing would perhaps be a combination of one of these and any other compact camera, whatever your needs are.

Just trying to think differently, while everybody seem to be waiting for Godot (aka DMD).

Stephen, the main text mentions visible smoothing artifacts at iso100; that is a pretty good clue that iso800 is not going to look unprocessed and clean.

And yes, a kit lens is typically f/3.5 or slower, which really is too slow for night-time even in a city. That's one reason I never bought it, and got a couple of fast lenses instead. No wonder, of course; the ability to change lenses is one of the benefits for which we accept large, bulky, expensive bodies in the first place.

There is a place for a small and light take-everywhere camera, just as there is a place for a DSLR (and just as there will be a place for large-format film cameras in the foreseeable future). They have different strenghts and different weaknesses - and this is a good thing. It's really shortchanging the compact camera if we try to shoehorn it into a role for which it doesn't fit and for which it was never designed.

Until the ideal camera comes out we have to be content with what is available. Waiting for something that may never be built isn't simply frustrating, it's counter productive. It's easy to find the faults and shortcomings of the vast selection of comapact cameras, we just have to weed through all the info and find what is best suited for us, buy it and start taking pictures. I just bought a Richo GRD and am pleasantly suprized. We should also let the manufacturers know what we want and maybe someday....

Fuji is infuriating ! For a couple years now, they've had this wonderful sensor that could very well be used to make the ultimate low light compact (well, given that Mike & Thom's large sensor compact just ain't happening !) So they put it in a pocket-cam with no OVF, a slow lens and no stabilization. After a couple years of "evolution" they still have no OVF, a slow lens; they've added stabilization and doubled the resolution. Ay yi yi !

"Canon digicams (including the mildly controversial but certainly able G7) employ the other leading form of mechanical VR, shifting elements within the lens instead of moving the sensor. Which is better?"

Sensor vs lens IS (I am a Canon user...) does not matter if you have a non-removable lens. One can only hope that the engineers have done their job and matched the IS to the lens to get the specified stabilization.

It only really matters with SLR's and removable lenses. The in-lens components can be placed in a spot where small movements in IS elements make *large* improvements in IS performance. Sensor based IS does not have that luxury.

Actually, I think that sensor based IS systems should state improvements in speed vs lens length, i.e., that the sensor based IS has no real improvement with 300 mm or above lenses (or whatever it is).

We must remember that noise which looks the same at 100% on the screen will be much less noticeable when you see the whole image, at higher resolution. Because each pixel will be printed or viewed much smaller. In other words, if a twelve megapixel camera is a bit noisy, it will matter less than if a six megapixel camera is.

A factor that's often overlooked when accessing the potential image quality of cameras having high pixel counts on such small sensors is that the enlargement factors suffered when trying to use all those pixels can induce visible degradation caused by diffraction. This directly inhibits the camera's ability to record subject detail at equivalent resolutions (in lp/mm) with all but the widest apertures.

The Fuji FinePix F50fd has a 1/1.6" sensor - having a width of about 8.2mm) but its 12 MP sensor encourages us to make 15.75 x 11.8-inch prints at 254 dpi. (4000 pixels / 254 = 15.8). We can be sure that many people make even larger prints with 12 MP cameras, but for the sake of this discussion let's assume the enlargement factor will be no greater than that suffered when the 4000x3000 pixel files are printed at 254 dpi: 49.8x. (This was calculated by dividing the diagonal of the print (19.7 inches) by the diameter of the sensor (10.038mm).

Equipped with the enlargement factor and a desired print resolution, we can calculate the f-Number at which diffraction would inhibit the desired resolution. 254 dpi equates to 5 lp/mm (254 dpi / 25.4 mm per inch / 2 lines per pair = 5 lp/mm):

N = 1 / enlargement factor / desired print resolution / 0.00135383

In this case:

N = 1 / 49.8 / 5 / 0.00135383 = 3.8

So, if we really want to secure 5 lp/mm worth of subject detail in a final print that's made from a sensor that requires a 49.8x enlargement factor at 254 dpi, we have to stop down NO FURTHER than f/3.8. The FinePix F50fd's maximum aperture is f/2.8 and can be set as small as f/8, so diffraction will prevent us from achieving 5 lp/mm in a 254 dpi-sized print made from this tiny sensor's 4000 x 3000 pixel files if we shoot at apertures smaller than f/4. (We can’t use f.5.6 or f/8 without reducing print resolution to something proportionately less than the 5 lp/mm had at f/3.8)

At a 49.8x enlargement factor, the camera's aperture range of f/2.8 through f/8 translates to resolutions of 5.3 lp/mm through 1.9 lp/mm, respectively. Thus, when shooting the Fuji FinePix F50fd at f/8, you'd have to increase your viewing distance from 10 inches (where 5 lp/mm can easily be appreciated by people with healthy eyes) to a minimum of 26.3 inches to accommodate the lack of subject detail rendered in a 48.8x enlargement using f/8.

I'll spare you the math, but if you want to use this camera to produce a print that literally possesses subject detail at the limits of what average healthy human eyes can appreciate at viewing distances as close as 10 inches (a desired print resolution of 8 lp/mm), you must limit your enlargement factor to 33.0x at f/2.8 or 11.5x at f/8. These enlargement factors translate to print dimensions of 10.5 x 7.75-inches for f/2.8, and only 3.7 x 2.7-inches for f/8. Ouch! Translation: The sensor is too small to make good use of all those pixels.

No amount of uprezzing can replace subject detail lost to diffraction before the light even struck the sensor. You can create counterfeit detail with products like Genuine Fractals, and you can certainly improve the acutance (edge sharpness) of diffraction-damaged images, but there’s no way to replace true subject detail (resolution) that was destroyed by diffraction.

I just can't get excited about this camera.

Mike Davis

Mike - thanks for the math - I knew that more & more high res cameras are becoming diffraction-limited, even wide open for part of their zoom range; the Fuji's use a bigger-than-most sensor, but I imagined 12MP had to push it over the limit. You mention the max aperture of f/2.8, but it's f/2.8-f/5.1 so if your math is correct, it's diffraction-limited over part of its range, even wide open. As someone else mentioned, noise gets averaged out in downsampling, but if you're avoiding large prints because of noise and because you're diffraction-limited on sharpness, then what's the point of a 12MP sensor ? Maybe it's not a big step down, but why not leave well enough alone ?

Mike Davis has offered a thought-provoking (if a bit numbers-heavy) response to my article. So I came up with a simple (or simple-minded?) test of my own.

I took the Fuji sample shot of the girl and the bubbles (see links in the original post) and did what I would do to any compact or dSLR capture. I added a bit of sharpening, and boosted the mid-tone contrast. (You don't have to do this at home, folks, you can use the straight-from-the-camera sample.)

Next, I resized the image without resampling, to 15 x 20". That gives us a resolution of 200 dpi. (No pixels added or subtracted.)

Now, my Epson R2400, modest printer that it is, only accomodates 13 x 19" paper, but no worries. For this test, I simply cropped the image to include the eye of the little girl that is closest to us in the picture. I printed it out on Moab Entrada matte paper. Drum roll, please...

Honestly, I was a bit shocked. Shocked at how good it looked! Viewed at a mere 9" distance under a high-intensity lamp. And this is the detail in a 15 x 20" enlargement, using the pixels captured by the camera, not a Genuine Fractals pretender!

Okay, kids, run and get out your printers...

Great, thanks, Stephen.

Indeed, the bigger the print, the less the pixel density needs to be, because viewing distance increases. I find 300 Pixels Per Inch (not DPI as it's often called) is only useful for small prints. For 8x10 to 15x20, 200 PPI is plenty. And for bigger, less will do.

What I'm missing in Mike explanation is where 0.00135383 comes from, and also, ALL THAT where comes from. Who determined this way of measuring it?

Fuji's sample 3 was taken with f2.8, which was consistent with Mike's calculations. Sample 1 was taken with f8.0.

Albano García:


(replying to Mike Davis...)

Your numbers are interesting so far as they go, but you didn't go far enough (and, for what it's worth, trying to take any of this to more than one significant figure is an exercise in wishful thinking).

First, the resolution of a Bayer array camera doesn't equal the pixel count. A typical camera today resolves about 40% of the pixel count. So, a 12 MP camera will typically have 5MP worth of real resolution. Or, to put it another way, the smallest useful blur circle is about 50% bigger than what you estimate it to be for a diffraction-limited f-number of 5.6. Not so unreasonable for a lens whose maximum aperture varies from f/2.8 to f/8.

BTW, this is a gross approximation-- computing the blur circle size that optimized sharpness for a Bayer array is a non-trivial proposition!

Second, your criterion for a sharp picture has little to do with what viewers think a sharp print is. There's ample test data that shows that discerning viewers perceive an 8x10 (not 12x16) inch print at close distance as being sharp when it renders about 4 lp/mm -- some say 3, some say 5, I say let's not split that hair too finely. That works out to about 2K by 2K pixels worth of real detail (Bayer resolution is not isotropic, so precise format ratios hardly matter). This is typical of what compact-sensor and low-end DSLR cameras are turning up in tests today, regardless of sensor format.

Third, your estimate of the "the limits of what average healthy human eyes can appreciate..." in terms of spatial detail is wrong. At closest comfortable focusing distance (18" for most humans, not 10", unless you're very near-sighted like me, in which case it's 5"), the resolution limit is indeed given variously as somewhere between 8 and 12 lp/mm (more imprecision!). BUT... that is not anywhere close to the limit of the human visual system, which will perceive fine detail up to about 30 lp/mm!

In other words, if you were to make up a series of matched 8x10 prints portraying resolutions of 1,2,4,8,16,32,and 64 lp/mm and ask discerning viewers to tell you which ones were sharp and which ones were sharpest, you'd get two very different answers. They'd discard the 1 and 2 lp/mm prints as unsharp, and everything above that would be considered sharp. BUT... they'd be able to perceive the 8 lp/mm print as sharper than the 4, and the 16 sharper than the 8 and the 32 sharper than the 16. They'd see no difference between the 32 and 64 lp/mm prints, though. 30 (roughly) really is the limit for perfect sharpness.

But everything between 4 and 32 lp/mm-- "sharp", for some definition of the term. There's nothing special about 8 lp/mm, it's just a step along the perceptual way. But if you're really seeking your perfect sharpness, you have to aim for 32 lp/mm. Good luck with that!

Fourth-- let's not forget the printer. A venerable workhorse like the Epson 2200 can't render more than 8-9 lp/mm in a print. Newer printers can do better, but I've yet to test anything that gets much better than 15 lp/mm. So that mystical 30 lp/mm print is out of reach. If you want perfection, it's back to contact-printing in the darkroom (you'll never get it in an enlargement).

pax / Ctein

Word salad crept into my typing-- the fourth sentence, which read:

"Or, to put it another way, the smallest useful blur circle is about 50% bigger than what you estimate it to be for a diffraction-limited f-number of 5.6."

should have read

"Or, to put it another way, the smallest useful blur circle is about 50% bigger than what you estimate it to be. That gives you a diffraction-limited f-number of 5.6 instead of 4."

Editing late at night... always a perilous procedure.

pax / Ctein

Dear Albano,

Here's a more intuitive way of looking at it. Mike Davis wants a 50X enlargement that resolves 8 lp/mm. That means the taking lens must resolve 8 x 50 = 400 lp/mm.

Handy formula for the diffraction-limited resolution of a lens is

aperture = 1500/resolution

where the aperture is the f-number and resolution is in lp/mm.

so, the taking lens in this case must have an aperture of at least:

1500/400 = f/4.

Which is what Mike got (here again I'm reminding folks that none of this deserves more precision than that).

If you'd be happy with an 8x10 print (about 30X) and 4 lp/mm (typical minimum value that discerning viewers would call "sharp"), then you need a taking lens that resolves:

120 lp/mm (4 x 30)

and that means your aperture must be at least:

1500/120 = f/11 to the nearest stop.

On the other hand, if you wanted a *perfectly* sharp 8x10, you'd be talking 30 lp/mm at 30X = 900 lp/mm from the lens, and that's about f/1.7!

(Good luck buying a camera lens that's diffraction-limited at f/1.7, by the way .)

pax / Ctein


I'm not surprised that you are more than satisfied with the results seen in your 200 ppi test, where the enlargement factor would be 63.3x vs. the 49.8x enlargement factor discussed in my 5 lp/mm example calculations.

At f/2.8, where the effects of diffraction would be the least destructive across the range of apertures offered by this camera, an enlargement factor of 63.3x would limit true subject detail at the print to:

lp/mm = 1 / enlargement factor / N / 0.00135383

1 / 63.3 / 2.8 / 0.00135383 = 4.2 lp/mm

That's not bad at all. Many people would agree that this is plenty of detail - even people who can actually discern the difference between acutance and resolution.

But at f/8, with a 63.3x enlargement factor, diffraction alone would prevent you from achieving a print resolution any higher than:

1 / 63.3 / 8 / 0.00135383 = 1.5 lp/mm

For those who can examine a print at a viewing distance of 10 inches, 1.5 lp/mm falls short of what many people can appreciate. Such a print can be sharpened, to improve acutance, but in terms of subject detail, it would be best viewed at a distance of 20 inches or more.



You may have already found this by following Matthew's suggestion, but here's a good discussion that makes reference to the constant 0.00135383:




As Stephen wrote, my post was "a bit numbers-heavy" without getting into the effect of the Bayer algorithm and anti-aliasing filters on system resolution, but it actually has no bearing on my argument. You've basically said that if we're not using a Foveon sensor, we can lower our expectations to 50% of the resolution that pixel math alone would say we can achieve. I've always assumed that the Bayer algorithm and anti-aliasing only rob us of about 30%, not 50% but, again, it's irrelevant because the size of the Airy disk after enlargement is not affected by any other limitations on total system resolution - plain and simple. Between us we could surely come up with 50 different factors other than diffraction that can reduce the realistic maximum achievable print resolution, but not one of them will have any bearing on the fact that after a given enlargement factor is imposed, the Airy disk size in the print will be exactly that many times larger than it was in the camera. That's true whether we've got a digital sensor at the image plane (Foveon, CCD, CMOS, etc.), or film, or a wet plate collodion emulsion. There are only three factors affecting the size of the Airy disk at the print: Enlargement factor, f-Number, and the color (wavelength) of the light. The latter variable is incorporated in the aforementioned constant, 0.00135383, which assumes we’re working with a wavelength of 555nm, in the yellow-green portion (in the middle) of the visible light spectrum.

Regarding your comment about what the average person perceives as a sharp print, I would like to reiterate that the "desired print resolution" used in my formula is a variable - not a constant. I chose 5 lp/mm for one example and 8 lp/mm for another, but my goal here was to equip readers with a tool they can use to determine the f-Number at which diffraction will begin to inhibit a desired print resolution of their choice for an anticipated enlargement factor. That choice should be made with consideration of the anticipated viewing distance as well as any subjective preferences, all under a ceiling of what resolution can realistically be achieved on-print, given the other factors affecting total system resolution. (Controlling diffraction is but one link in the chain.)

Print resolutions far higher than 8 lp/mm might well be appreciated, but In John B. Williams' book, "Image Clarity - High Resolution Photography", he writes that although resolutions as high as 25 lp/mm can be perceived by some people, there's a shoulder that begins to fall off fairly steeply at around 8 lp/mm. Neither you nor I have any intention of limiting our print dimensions to support a "desired" print resolution of 25 lp/mm, but I agree with your comment and am both aware of and grateful for the considerable research you've done in this area. Your paragraph, above, that begins with "In other words..." tells me that we're in complete agreement as to what one should consider as a realistic "desired print resolution" at minimum possible viewing distances.

Lastly, I must respectfully point out that you misquoted me when you wrote: "Mike Davis wants a 50X enlargement that resolves 8 lp/mm." Nowhere in my posts to this thread have I stated what my preferences are, either for enlargement factor or print resolution. If anything, the most that one could infer from reading my first post to this thread is that I have a preference for enlargement factors far less than 50x (that I prefer larger sensors and/or smaller prints) and, perhaps, that I consider a desired print resolution of 5 lp/mm (not 8 lp/mm) to be sufficient when a viewing distance of 10 inches is anticipated.


Mike Davis

Dear Mike,

Yeah, we're substantially in agreement.

BTW, when I said "Mike wants..." I was just speaking colloquially. Didn't mean to really imply I knew what was in your head.

Do you really mean a 10" viewing distance? That's damn close-- closer than most humans can focus.

Minor correction-- a typical Bayer array camera performs about as well as if it had a true-pixel count of 40% of that. The linear resolution only drops by the square root of the pixel count. Hence, linear resolution is about 65% of what you'd get from a full-res sensor. So, your average mental rule of a 30% resolution loss is correct!

FWIW, there have been Bayer array cameras and backs out there that resolved as poorly as 25% of pixel count (that gives you that 50% linear resolution loss). At least one, though, managed 65% of pixel count (a mere 20% resolution loss). I'm not sure how they pulled that off; I'd have guessed that was theoretically impossible. Seems not. Almost every current camera falls between 33 and 50%.

Regarding limiting resolutions, I'd go one further than Williams, even-- I'd say the effects of improvements drop off pretty fast above even 5 lp/mm. And the jumps are proportional. That is, people will report a big difference between, say, 2.5 and 5 lp/mm prints, but not so big between 5 and 10. And the difference between 5 and 10 is at least as prominent as between 10 and 20.

A point worth noting-- except under unusual circumstances, people will never report the 5, 10 and 20 lp/mm prints as being unsharp when viewed in isolation. It's only when put side-by-side that discerning viewers can readily pick out the sharper print. One of those cases of "better" being the enemy of "good."

In the pursuit of objectivity, I should mention that I've heard a counter-argument to the position you, I, and Williams take that high-resolution numbers are perceivable. The counter is that generally there's a rolloff in the MTF well before the cutoff frequency, so it's possible that we're not directly perceiving the higher frequencies but only the relative contrast at lower (5-8 lp/mm) frequencies.

I strongly *think* that's wrong, but I've yet to come up with a compelling experiment that I could do that would prove that. So it remains a possibility.

pax / Ctein

I don't mean to butt in here, but just a few words about Ctein's point about comparative differences being more discernible: when I did my big print comparison test for the old C&D, we showed more than 30 people, including 6 photographers, 11x14" prints from both 4x5 and 6x7 negs. All the photographers could pick out the 4x5 print once they knew what they were looking for; only a small minority of the non-photographers could (I think it was 4 or 5 out of the 24). The rest said they couldn't see a difference. At 16x20 the differences between the two film formats became obvious to almost everybody.

But the thing I remembered in this context is the photographers going back and forth, back and forth from print to print, squinting first at one and then immediately at the other.

The same thing happened to me when I took a color test at NIMH. I was able to line up discs correctly according to slight variances in hue, value, saturation, etc., whereas that ability is very rare in the untrained population. I thought it was merely due to the fact that I was a color printer and worked at discerning those kinds of differences all day long. (I'll have to ask my brother, but I think the researchers began to disqualify color printers as research subjects after my visit!)

A consideration this brings up is that how viewers have been trained to discern differences will also probably factor into how well they can see resolution differences, comparatively or not. Experimenters might need to differentiate between graphics professionals and untrained members of the general public when interpreting their data!

Also, contrast at lower frequencies becomes a big issue when evaluating lenses, and I personally think that peoples' perception is different for B&W than it is for color. High contrast at 5 lp/mm looks VERY sharp to me with B&W film even when it is confirmed to lack very fine detail detail under the loupe (I have some very interesting test prints, and it's fascinating to look at MTF charts alongside prints made with the lenses you have charts for). But my impression of such a lens isn't as good when it's used with color film.

I have no idea if this last is just my personal impression or if it would also correlate with what other people see.



Thanks for clarifying your position. I'm glad we're in agreement, believe me. This may sound overly modest, but I genuinely respect your extensive knowledge of this topic.

Yes, I use 10 inches as the distance at which a print might suffer its greatest scrutiny. I can't tailor my prints to satisfy those who are severely myopic (and certainly not those who use loupes or magnifying glasses), but I've read many references which say that "healthy eyes" can focus as close as 10 inches. I personally cannot focus any closer than 12 inches.

I can't exaggerate my glee at reading your comment regarding side-by-side comparisons. I have many times argued, seemingly to deaf ears, that it is precisely because diffraction affects the entire image uniformly that it goes insidiously undetected by those who have no discernment for it. It's comparatively VERY easy to detect defocus in a standalone print. When the circles of confusion at the Near and Far sharps of the subject space are "detectably" larger than those which lie closer to the plane of sharpest focus, anybody can say, "Gee, this part of the print isn't as sharp as that part of the print." But when the resolution of the entire print has been reduced uniformly by diffraction, even to a degree that would be easily detectable in a side-by-side comparison, it goes unnoticed by most people.

Add to this phenomenon the fact that most people can't assess the difference between acutance and resolution. A paint-by-numbers painting can be executed with a fastidious attention to edge sharpness, but even the most precise effort will still deliver very limited subject detail. I see many sharpened prints that are, to my eye, little more than number paintings. It’s very common to see prints having high acutance, but low resolution, and exceedingly rare to see a print that possesses both. I crave detail. At whatever viewing distance is appropriate for a print of a given size, I want to see detail that’s right at the limit of what my eyes can perceive. Ideally, I want to walk right up to the print (to a distance of 12 inches in my case) and experience no awareness whatsoever of the building blocks of the image.

This takes us to Mike's comments. In general, I agree completely that the average person, who I like to call ‘Joe Consumer’, simply has very little ability to discern image quality. If we want to sell an image to Joe Consumer, we can get away with far less resolution than we ourselves find acceptable. The same thinking holds true if we want to sell a camera to Joe Consumer. It's Joe Consumer's inability to appreciate image clarity that's allowing manufacturers to jam so many pixels onto such tiny, inexpensive sensors, with proportionately small, slow, and relatively inexpensive lenses. The market has spoken and will continue to speak. Those of us who understand the need for a camera like the Sigma DP1 (and better still), are among a very small minority who will never get the ear of the industry as a whole.

Mike Davis

Mike D.,
I'm having some difficulty with some of your comments in your last post. Are you saying that standard technique DOESN'T give you enough resolution? Some of the digital images I look at can resolve MAKEUP. Next, you seem to be saying that pictures SHOULD resolve more. But any technical property is just that--a property--not a value. As an example, look no further than the Steichen autochrome I just published. Finally, I've always found "Joe Consumer's" take on image quality to be a good "brake" on photo-techie obsessions. I always listen carefully to non-specialists, because I want to know how they're perceiving a photograph. Here's a small passage from an essay of mine called "Lens Tests Are Always Wrong": "Most people don’t look at optical quality when they look at pictures. They look at the pictures. In this sense they may have a big advantage over the typical photographer. Gazing appreciatively at a print, a proud photographer may be thinking, 'look at that superb edge contrast between the red and the green--and what beautiful gradations in the reds!' while his non-photographer friend standing next to him is thinking, 'That’s a geranium.'"

Mike J.

This discussion has certainly had legs. And a growing pile of numbers, too. Regarding print sizes (and a smidgeon on viewing distance), check out the TOP article from Thursday, October 19th, 2006, "Burt Keppler on Megapixels".

I find it amusing (and instructive) to look at the delicious autochrome by Steichen that Mike posted yesterday (7/31/07). Don't you just want to drag it into Photoshop and hit the Autocolor stick? Or reshoot altogether so that the damned thing is sharper? And so we can get some decent shadow detail? Not.

It is one thing to be able to make a picture like the Autochrome as an artistic choice. It is quite another to be forced to get a picture like that because of your tools.

Even though Mike Davis strives for the most resolution he can get, he can still make the choice to smear vaseline on the lens and get a soft image if that is what he wants. However, if the whole system is soft, there is no way to get a sharp, clear image even if that is what the artist wants.

Dear Stephen,

Well about that Autochrome...

Putting aside the question of what it really was supposed to look like (I've never seen one in pristine condition, so I have no idea if the yellowing and desaturation are original or the result of deterioration)...

My artistic reaction to it is reminiscent of that hilarious Nikon ad of some decades ago that showed the infamous phony Loch Ness Monster photo and read (I'm paraphrasing, 'cause I don't remember the precise wording):

"Some people see a monster.

We see poor composition, bad exposure, and a lack of focus."

My gut reaction as an artist (not a technician) to that Autochrome is that it's a great antique and (in its present condition) a lousy photograph.

To each his own, eh?

pax / Ctein

Dear Mike D,

Whoa there, fella, ... we've not agreement on your primary argument, which is that small sensors are compromising image sharpness. I definitely don't support that position.

For one, as I showed, diffraction ain't gonna matter until you're zoomed out to the extreme telephoto end of the lens. I can promise you there's no way that lens performs at its diffraction limit at max zoom and f/8 (neither, for that matter, do regular camera lenses in those long focal lengths). So arguments about diffraction are pointless.

For two, tests of current high end compacts and DSLRs show them producing about the same amount of in-camera resolution. The smaller sensors simply are not compromising sharpness. Maybe with better, future cameras they will, but right now it's not of any import. Kind of like old arguments I'd read about whether TMAX 100 or Tech Pan got you sharper medium and large-format photos (answer-- it weren't the film that was the limit factor).

As for the future, I think we will eventually have full-color-pixels as the norm (though not necessarily the Foveon technology). I should not want to attempt to put a date on it, nor would I bet much money. It's my hunch, though.

Now, if we could just get Sigma/Foveon to stop their deceptive and dishonest advertising...

pax / Ctein


You've really startled me with your latest post. I'm frankly astonished that you could write, "So arguments about diffraction are pointless." Your premise for this was, "I can promise you there's no way that lens performs at its diffraction limit at max zoom and f/8 (neither, for that matter, do regular camera lenses in those long focal lengths)."

I’ve not been calculating the f-Number at which the lens will perform "at its diffraction limit." It would be wonderful if diffraction could not degrade our images until the diffraction limit of the lens was reached, but that's simply not the case.

The following equation (provided earlier) tells us the f-Number at which diffraction's Airy disk will reach a diameter that begins to inhibit a desired print resolution at an anticipated enlargement factor:

N = 1 / enlargement factor / desired print resolution / 0.00135383

The formula tells us when we will be able to "see" diffraction's degredation in the final print. If the enlargement factor is great enough, the print could suffer visible degradation due to diffraction even when shooting wide open.

Your reference to "tests of current high end compacts and DSLRs [which] show them producing about the same amount of in-camera resolution" is frustratingly unqualified, but you offer it in a way that suggests we can rest easy and forever ignore these undisputable facts:

1) Small sensors require greater enlargement factors than large sensors to reach a given print size.


2) The diameter of Airy disks at the sensor plane are determined by the f-Number used and the wavelength of light.

From these two facts alone, without using any math, you should be able to see that at any given f-Number, the Airy disk diameters found in a print of a given size, produced from the 12-Megapixel Fuji FinePix F50fd, will be LARGER than the Airy disk diameters found in a print of the same size, produced from a 12-Megapixel full frame sensor.

Hint: The Airy disks will be the same size at the sensor of each camera, but the Fuji FinePix F50fd will require an enlargement factor that's a whopping 4.3 times greater than that required by the full frame sensor to make like-sized prints, and thus, after enlargement, the Fuji’s Airy disks will be 4.3 times larger at a given f-Number. The Fuji will be 4.3 times more likely to produce Airy disks that can inhibit a desired print resolution.

Whenever we impose a combination of enlargement factor and f-Number that forces Airy disk diameters at the print to exceed about 1/5 to 1/2 a millimeter, discussions of diffraction become anything but "pointless.” Many 10- and 12-megapixel digicam users are imposing just such combinations of enlargement factor and f-Number when they try to produce even 254 ppi-dimensioned prints from those tiny, high-density sensors.

Are they happy with the results? Of course they are. If they weren’t, the manufacturers could not possibly continue to up “the ante” as Stephen put it in his original article. But we're talking about the same people who are satisfied with their enormous flat screen HDTV's having vertical resolutions as low as 480 pixels. Joe Consumer’s contentment does not have any influence on the physics of diffraction, nor on my desire to equip people with a simple formula to determine when diffraction will inhibit the print resolutions THEY hope to achieve.

I hope you will carefully reconsider the material I'm presenting here. Your words carry a lot of weight.

Thank you,

Mike Davis

Yeah Ctein, what's up with that? Even if diffraction isn't the dominant source of degradation, it's still *a* source of degradation in images that might also be affected by residual amounts of other aberrations, right? And diffraction should theoretically be a much bigger issue with tiny sensors than with large ones.


Dear Mike(s),

Yeah, ... " diffraction should theoretically be a much bigger issue with tiny sensors than with large ones."

And, as I pointed out by way of analogy, film resolution should be a big issue... 'cept it's so often not.

When other factors dominate the performance, diffraction effects (or film resolution) don't count for much.

Proof's in the pudding. As I said, you don't see big sensor cameras massively outperforming the high-end compact small-sensor cameras. I didn't qualify the statement because it needed no qualification-- it's an across-the-board observation.

But ya want specifics? OK, I pulled several recent issues of Pop Photo and grabbed the camera test resolutions. Here's pixel count, sensor size (relative to a 35mm frame) and measured resolution in lines.

7 MP 1/6x 1700 lines
8 MP 1/2x 1800 lines
10 MP 1/6x 2000 lines
10 MP 1/5x 2100 lines
10 MP 2/3x 2100 lines

There was no recent full-frame-sensor test (you can't cmpare data across test procedures that have changed with time). But this makes the point. There's no significant correlation between sensor size and resolution in today's camera's, up to 10 MP.

What amazes me is how closely resolution correlates with pure pixel count, considering these are cameras from four different manufacturers, of very different designs. But it's not a fluke-- I've observed this over many, many cameras.

I am certain 12 MP will produce similar results-- that's only a 10% change in pixel size. Not enough to change the physics nor the dominant factors, whatever they may prove to be. Diffraction ain't gonna be one of them, I guarantee it.

Tomorrow's cameras? Who knows! Contrary to Mike D's remarks, I did not say we could "forever" ignore anything. I was very careful to couch my remarks in current performance. We can for now ignore it.

With present cameras, dithering over diffraction limits has as much import as dithering over film resolutions did in large-format photography. That is, pretty much none.

pax / Ctein

Forest for the trees, guys?

Exactly what are we using this and similar cameras for?

This camera, we now know, can produce fairly large prints with excellent detail in good lighting conditions, albeit with some noise and some artifacts. So far, so good.

How it will perform in low light conditions is anybody's guess at this point (although unfortunatly it appears it will not come close to older f-series models).

Personally, I am not so concerned with the diffraction issue being argued here so fervently. Why? Two reasons.

First, these cams rarely *need* to be used at higher f-stops. They give a lot of depth of field at large apertures.

Secondly, I would argue that lens diffraction at higher f-stops *can* be used artistically. I daresay that diffraction as well as exposure compensation manipulations are about the only artistic ploys these cams offer, unless you happen to like the digital noise and/or the NR smearing effect at high ISO.

Finally, Fuji has a habit of "backfilling" its product lineup. I'll bet we will soon see the f41, etc, upgraded with IS soon. :)


The proof may be in the pudding, but again, you're presenting unqualified data to support your argument. We're talking about diffraction here, so it's entirely reasonable for me to ask: At what apertures were those Pop Photo resolutions achieved?

In my first post I made no effort to hide my belief that an F50fd user can prevent diffraction from inhibiting a desired print resolution of 5 lp/mm in a 254-dpi dimensioned print by shooting at f/3.8 or wider, so you're not rebutting anything I've said by quoting test results that lack any reference to the apertures used. (It's not as if I've been saying that these high-density sensors can't produce good results at ANY combination of f-Number and enlargement factor.)

The drum I'm beating (or is it a dead horse?) warns that the smaller apertures available on digicams equipped with these high-density sensors should be avoided when the intent is to make a print as large as all those pixels would encourage us to make -AND- I've provided a formula for determining what f-Numbers must be avoided to support a desired print resolution at any anticipated enlargement factor.

But you seem to be promoting the idea that we can completely ignore the effects of diffraction, at any and all combinations of enlargement factor and f-Number we're likely to encounter when using any digital camera. If that's not the case, could you please define the limits of your optimism regarding diffraction?

Back to your pudding... Are the quoted results for vertical lines of resolution or horizontal lines? Assuming they are horizontal lines, I'm certain those results (averaging about 2000 lines) can not be achieved at f/8 using the smallest sensors because diffraction alone would limit resolution to 92.3 line pairs per mm (185 lines per mm), assuming a wavelength of 555nm. This is true for any format.

The third example you provided shows a 10 Megapixel 1/6x sensor achieving 2000 lines. That 1/6x sensor has a height of only 4 mm (that's 1/6th of a full frame sensor's 24mm height), and thus, at f/8, diffraction would limit the resolution to 4 * 185 = 740 lines. 740 lines is well short of the 2000 lines you're saying a 1/6x sensor has achieved at an *unspecified* aperture. Diffraction would require the user to open up to f/4 to achieve 1480 lines across the 4 mm height of a 1/6x sensor, and 2000 lines wouldn't be possible until the user opened up to f/3.0, so again I ask: At what apertures were those Pop Photo resolutions achieved?

Give me the pudding.

Mike Davis

Dear Mike,

Sorry, no. You're the one who has yet to produce any pudding. You've produced a theoretical analysis; I've cited real-world data. If you're not convinced, that's not my problem; I think the readership is.

Your f/8 at 1/6 X is a red herring. A very rough but useful rule of thumb is that to maintain constant depth of field, the aperture has to scale with the format size. That is, take a bunch of different format cameras, equipped with lenses of proportional focal length so the field of coverage is the same for all of them. Make same-size (not same magnification) prints from all the formats. How do you have to set the lens apertures to get the same depth of field in all the photos? The extremely crude answer is that the aperture will scale with size.

In other words, if f/5.6 gives you enough depth of field in a 35mm neg, you need to go to f/11 for 6x7 cm, f/19 for 4x5 inch and f/4 for 8x10 inch formats.

In the same vein, a lens stopped down to f/8 with a 1/6X sensor is providing the same depth of field as a (proportionally bigger) full frame sensor lens stopped down to f/48. The import of that:

First is that you'd never stop down that far by preference. You may be stuck there with a slow lens, but that's a different matter. The slow lens won't perform diffraction limited wide-open anyway. Second is that neither format will then meet criteria for critical sharpness; f/48 is simply too small an aperture and diffraction effects WILL dominate both.

You started out by simply dismissing the camera for its small sensor. Do not chastise me for making unqualified statements. Mine require considerably less qualification than yours.

You need to go do some real-world research. Lots of online data to answer the questions you're asking of me. I'll give you one tidbit-- today's Bayer array cameras typically produce the about same number of line-pairs of resolution horizontally, vertically, and diagonally. Blur circle is no longer a simple concept. So maybe you need to use 4mm in your 1/6x calcs... or maybe 6 mm... or maybe the 8 mm diagonal?

Beats me. Figure it out. Get your equations to match with the REAL WORLD comparative data, instead of the other way around. Then we'll have something more to talk about.

Until then, I feel I've made my case more than adequately. If you don't feel you have done the same for yours, feel free to claim the last word. I've nothing more to add.

pax / Ctein

The comments to this entry are closed.



Blog powered by Typepad
Member since 06/2007