Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: Gary Ferguson on February 19, 2006, 06:46:12 am
-
As a darkroom printer I arrived at certain guidelines for enlargements. They weren't rules as such, but observations that tended to fit with my own experiences.
4x or 5x enlargement was not difficult, undemanding of equipment and technique. This degree of enlargement served up both the 6"x4" holiday snaps from 35mm, or delivered a magnificent 16"x20" print from a more carefully executed 4"x5" negative.
7x or 8x enlargement was entirely possible, and reliably so under both good conditions and bad. But it did require that some attention be paid to technique, and it did start to show the differences between good and average equipment. This was the normal "enthusiast's enlargement", yielding the de facto 35mm standard 8"x10" print, or a 16"x20" from a medium format negative.
10x enlargement was ambitious. This was time to take a deep breath and concentrate because most prints at this level of enlargement would compare unfavourably with ones from the next larger negative size. 10x was printing on 11"x14" paper from a 35mm negative, and you'd usually see straight away that it wasn't medium format. It could still be a convincing print that withstood close inspection, but only if every single link in the photographic chain was meticulously handled.
12x to 15x enlargement represented the absolute limit. Not only was equipment and technique stretched to breaking point, but you really needed the photographic gods to be smiling on you. The truth was that the great majority of negatives could not sustain this degree of enlargement without their technical shortcomings becoming manifest, and if you were successful at this scale it was likely to be serendipitous and unrepeatable. In other words, I may have made a print that was a 12x enlargement or even greater, but I didn't have that intention when I took the shot.
So what were the standards that drove these guidelines? Well, I never bought into the theory that viewing distances increase in proportion to print size. Apart from artificial restrictions like a poster innaccessibly displayed on the side of a building, it seemed to me that there's something in the very nature of a photograph that invites leisured, intimate scrutiny. And for many subjects, often found in landscape or architectural photography, abundant detail is necessary for the success of the image.
Now here's the thing. Personally I don't find much in digital photography that might change the enlargement guidelines of silver photography. Grain may be solved and film curl replaced by sensor alignement; but from diffraction to camera shake, depth of field to mirror slap, we're still dealing with the same practical restrictions on fidelity. Yet somehow in the digital world huge enlargements seem to be regarded as a pixel given birth right, so an A2 print from an APS sized sensor doesn't even raise an eyebrow.
I'm looking at the 11x prints I'm making from a Canon 1Ds Mk II and concluding that I'm already pushing my luck. I'm not sure what would be the practical benefits from a new 22MP Canon uber kamera that could be realised in a print.
So my question is this, what is the realistic relationship between sensor size (not pixel count) and print size?
How big is too big?
-
So my question is this, what is the realistic relationship between sensor size (not pixel count) and print size?
[a href=\"index.php?act=findpost&pid=58525\"][{POST_SNAPBACK}][/a]
As I understand, Gary, there's no direct relationship between sensor size and print size. It's the pixel count and lens quality that determines how big a print you can make, that will withstand close scrutiny. The 16 megapixels of the 1Ds2 are good for a 14"x21" print at 240ppi. You could also make a print of around 18"x28" using 180ppi and no interpolation, but there would be a risk that some folks with really keen eyesight might see a hint of jaggies, so it would probably be advisable to interpolate the image (add more pixels) so it becomes 240ppi at that size.
Film really is in a different ball park. You have grains and clumps of grains that unavoidably get enlarged in proportion to the print size. Pixels also get enlarged but you can create more of them through the process of interpolation.
-
It depends a lot also on how the photo will be used, how it was processed, and how far away viewers will stand.
To my certain knowledge billboard photos aren't restricted to shots from 11x14 negatives, to cite an almost-frivilous example of the extremes, yet image quality is sufficient to support a major ad industry.
-
Digital enlargement is a bit different to traditional film enlargement, because we can interpolate and sharpen digital images, and digital printers don't lose contrast on bigger prints. The best measure for digital print quality from a well-sharpened image seems to be the print dpi. An 11x enlargement from a 1Ds2 is equivalent to 320dpi. For top quality prints that may be about right, although I have always been happy with 280-300dpi.
Iain West
-
As I understand, Gary, there's no direct relationship between sensor size and print size. It's the pixel count and lens quality that determines how big a print you can make, that will withstand close scrutiny.
[a href=\"index.php?act=findpost&pid=58527\"][{POST_SNAPBACK}][/a]
Agreed: from everything I have seen and read, the ultimate enlargement limit of images from the Canon 1DsMkII is still sensor resolution, at least when one uses the best available lenses. And why not use the best lenses once you have payed almost $8,000 for the body?!
For example, the Canon 24-70 f/2.8 is said by Norm Koren to resolve about 61lp/mm at 50% MTF, and it takes about three Bayer pattern pixels to resolve one line pair, so that corresponds to about 180 pixels per mm, or 5.5 micron pixel spacing (as in the Nikon D2X). So sensor resolution will only match that lens' resolution at about 28MP in the 24x36mm format of the 1Ds family. And I would guess that a number of good primes resolve better than that zoom.
-
"For example, the Canon 24-70 f/2.8 is said by Norm Koren to resolve about 61lp/mm"
I'm not so sure. Staying with your argument, how many lp/pm would you want in the final print? For quality work there seems to be a consensus that about six lp/pm would be a reasonable number, personally I'd argue for more but I'll live with six. In which case the x10 enlargement limit I spoke about earlier would still apply. In other words the A3/A3+ paper size is about the most you could reasonably expect to reach from a 35mm sized sensor with careful technique.
However, the quality test that I'd advocate is sterner; I'd ask at what point does an enlargement start to look significantly inferior to one made from a larger negative? In the days of film IMO a 35mm could generally hold its own up to a 10"x8" print, for a really good negative you could push that up to an 11"x14". After which the challenge becomes almost impossible as you pushed further into medium format territory.
What I see in my prints (using a 5D, 1Ds Mk II, P25 with a Hasselblad, and a P25 on a Linhof movement camera stitched to a sensor equivilant size of 49mmx71mm) is that even though digital gives a little more enlargement capability than film, especially with longer prime lenses, it's not that much more. And why should it be?
To go back to the 28MP Canon you mentioned (and I agree that something like it is sure to arrive before too long), some will look at that pixel spec and conclude that 240 dpi will get them pretty close to a 20" x 30" print. From 35mm? Surely it's time for a sanity check. What would the practical depth of field constraints be for an 18x or 20x enlargement? What back-breakingly heavy tripod would be required to still every last micron of vibration? And if there's a growing sense of dissatisfaction now with Canon's wide angles, what would the feeling be when those same optics are scrutinised at massive enlargement scales?
-
I also disagree with my friend Norm's calculations. Not with the fact that the math may be right, but that in the real world the numbers don't correlate with experience, mine or that of other photographers whose judgement I respect. I'm talking about on prints.
Right now we are lens constrained in most cases, not sensor constrained.
Michael
-
Right now we are lens constrained in most cases, not sensor constrained.
Michael
[a href=\"index.php?act=findpost&pid=58625\"][{POST_SNAPBACK}][/a]
Michael, does that also apply to Canon's best primes? The comments of yours I can recall are about Canon's zooms, and maybe some if its wide angle primes, which apparently fall a bit short of the state of the art for 35mm format wides.
Not that this effects the big picture, which is that for a Canon SLR to give significantly more resolution that what is possible with the 1Ds is likely to restrict one to a small elite selection of their current lenses, and likely put pressure on Canon to add some new high end lenses.
-
To go back to the 28MP Canon you mentioned ... What would the practical depth of field constraints be for an 18x or 20x enlargement? What back-breakingly heavy tripod would be required to still every last micron of vibration? And if there's a growing sense of dissatisfaction now with Canon's wide angles, what would the feeling be when those same optics are scrutinised at massive enlargement scales?
[a href=\"index.php?act=findpost&pid=58624\"][{POST_SNAPBACK}][/a]
All good points! In particular, the greater magnification of Out Of Focus effects ("DOF") with greater enlargements progressively pushes one to higher f-stops, while the need to control diffraction pushes one to smaller f-stops, and at 28MP, these jaws might close in many situations, offering no aperture size that adequately balances the OOF diffraction effects on the sharpness/detail/resolution that 28MP is capable of.
By the way, this effects even larger formats like MF backs equally, just shifting the relevant aperture ratio about one stop higher. I am still waiting to see how those OOF/diffraction trade-offs play out with the new digital backs at 30MP and up.
-
"I am still waiting to see how those OOF/diffraction trade-offs play out with the new digital backs at 30MP and up"
I take your point. The 30MP+ backs are a little rich for my blood, but I'm using a 22MP back and I've concluded that the phrase, "there's no such thing as a free lunch" could have been coined for photography!
I'm increasingly using the digital back mounted on a large format/movement camera in order to try and control depth of field with camera movements rather than stopping down. However, even when the shot allows this strategy there's still the problem of aligning the back with sufficient precision, just a speck of lint or dust in the mounting can be enough to throw the results out.
Still, if it was easy I guess we'd all be doing something else!
-
Surely it's time for a sanity check. What would the practical depth of field constraints be for an 18x or 20x enlargement?
There are two aspects to DoF. A scientific one based upon absolute resolution (ie. lp/mm of different parts of the scene) and a subjective one based upon an 'impression' from an appropriate viewing distance.
A consequence of lens improvement is that the parts of the image that are in focus will be sharper, but the parts of the scene 'not in focus' will not be sharper. There is consequently a greater disparity between what's in focus and what's not in focus, and therefore an 'actual', lp/mm, scientific, shallower DoF.
This seems unavoidable. But I would question the sanity of anyone who expects out-of-focus parts of any image to be as sharp as in-focus parts. We all know there is only one plane of focus.
There are always technological means being developed to overcome such problems. If your pixels are too small to generate sufficient dynamic range, try filling and emptying them several times during one long exposure. If the scene itself is just too contrasty for even relatively large pixels, then employ AEB and blend the images in PS. If you have a conflict between DoF requirements and absolute resolution, ie. if you insist that 30x40" prints should have the same DoF close up as they do from a 'normal' viewing distance, then you will have to employ ADOFB (auto depth of field bracketing which hasn't been developed yet but no doubt will by the time we get 28MP 35mm sensors). There are already available programs that blend differently focussed images to create enormous DoF from the scientific lp/mm perspective. All that's required now is an automation of that process.
What back-breakingly heavy tripod would be required to still every last micron of vibration?
As BJL has implied, it's all to do with pixel count and FoV. As regards holding the camera steady, it makes little difference if we have 28mp on a 35mm sensors, a 2/3rds sensor 1/16th the area, or an MF sensor 4x the area. The camera has to be held equally steady for all formats, for the same FoV and pixel count. The sorts of things that will change the outcome, however, are vibrations from mirror slap and mechanical shutters. The EVF cameras with electronic shutters will have the advantage with extremely high pixel count sensors.
And if there's a growing sense of dissatisfaction now with Canon's wide angles, what would the feeling be when those same optics are scrutinised at massive enlargement scales?
It's always unfortunate when one aspect of technological development streaks ahead of another essential component in the system. One would hope that Canon's new lenses would be up to the task when they introduce their ultra-high pixel density FF 35mm cameras. Perhaps we might even be able to trust them not to release a 28mp body until they've designed an appropriate lens. Who knows! Perhaps it depends on how savvy they think the camera buying public is.
There's an interesting analogy with Hi Fi audio here. For a long time now, amplifiers have exceeded the specifications of loudspeakers. Your sound quality is basically limited by the quality of your loudspeakers and your living room acoustics. You can spend as much on a state-of-the-art amplifier as on a Canon 1Ds2 body (and more). But any improvement with your existing loudspeakers in the same living room would be less noticeably than the resolution improvement of a 20D over a D60.
It's the perrenial problem again. Things that have to interface with 'reality' (ie. outside controlled laboratory conditions) are really difficult to improve. There are too many variables.
I'm editing this post in order to have a liitle rant which I think is relevant. My latest computer was built about 6 months ago. A Pentium dual core 64 bit processor, 3 GB of RAM and 400GB of Hard Drive. It was supposed to be my main computer for editing and storing photos. It's a white elephant. My Xrite-DTP94 colormeter does not yet have a 64 bit driver. I cannot calibrate my Sony CRT monitor with my fastest and best computer because production of a 64 bit driver is too difficult for X-rite or ColorEyes. And this is essentially a laboratory situation; just the monitor, the video card and the software. Very disturbing!
-
Another thing which thus far has not been discussed is frame content and how that relates to enlargeability.
I'm fond of saying that for enlargement purposes film is grain limited while digital is resolution limited. So with film, no matter what the subject matter grain becomes the limiting factor usually long before resolution exhaustion. Because of the very clean nature of digital, it's quite possible at some point in the enlargement process to actually see the effects of resolution exhaustion.
This brings up the issue of frame content. Both frame content and subject impact enlargeability. In general, the more real world geography which is captured in a frame the more stringent the resolution requirements for a given print size. The amount of fine detail which presents itself to be defined in a hyperfocal, wide angle landscape verusus that in something like a head and shoulder portrait with an OOF background is markedly different. So the resolution requirements for a worst case scenario (finely detailed, hyperfocal, wide angle langscape) and a much better case scenario (head and shoulder portrait with OOF and virtually featureless background) differ accordingly.
I've seen excellent 50 inch prints of head and shoulders portraits made from a D30 (3 megapixel) capture and very marginal 16x24" prints of hyperfocal, wide ange, finely detailed landscaes made with my 1DS (11 megapixels).
To be overly simplistic - there is a finite amount of fine detail of any serious importance in a normal head and shoulders portrait. Eyes, lips, a small amount of visible hair, skin pores (not that we really want to even detail this). In a wide angle, hyperfocal landscape filled with vegetation, rocks, etc., there is by comparison nearly an infinite amount of detail to deal with.
Bottom line is that when using today's fine digital equipment we must be sensitive to not only the amount of magnification in the enlargement but also to the subject and the angle of view encompassed in the capture.
Best regards,
Lin
-
Lin, you make some valuable points, and I'd like to take them further.
You argue that some subjects have inherently more detail or information than others, and therefore require different levels of resolution to be successfully photographed.
I agree, but I believe that as well as the inherent detail within a scene, we also have an inherent set of expectations regarding the amount of detail or information we expect to see in a photograph. These expectations are based on our basic experiences of seeing the world around us.
For example, a macro photograph of a fly that shows the structure of the fly's eye is so astonishing that we'll forgive many technical shortcomings and still regard it as "sharp". Conversely, with many of the cityscapes/architectural shots that I take, the viewer has a clear shopping list of expectations. So, if there's wording on a distant street poster, even if rendered very small, they expect to be able to read it.
-
Do some relatively large, sharp photos reveal more detail than a person would discern when looking at the actual scene? For example, could a distant poster be legible in a photo, but not when viewed with an unaided eye from the same position where the photo was taken? If so, is this extra detail produced even with lenses wider than a "normal" lens?
I've not done any tests, but it seems to me that I often see details in some of my photos that I could not see when I took the photos. Maybe my memory is faulty or I just didn't notice such details when I took the photos, but I seem to see more detail in some photos even when shot with a fairly wide lens.
-
"Do some relatively large, sharp photos reveal more detail than a person would discern when looking at the actual scene?"
Great question! As you pointed out it depends on the focal length of the lens, but that's certainly what I'd aim to deliver to the viewer, it's what I'd call the "binocular effect". Have you ever noticed that if someone picks up a pair of binoculars they stare fascinated at objects that normally wouldn't entertain their attention?
I've noticed that when confronted with a scene in real life the brain fills in many of the details, but when confronted with the same scene in a photograph the viewer interogates it much more objectively and literally.
For example, I once did a series of rooftop views. While taking the shots I'd try and find the distance when I could no longer distinguish between concrete, brick, and rendered brick. The problem is that in real life you're influenced by other clues, does the building have a slate roof, or chimneys, is it located in a predominantly Victorian section of town? I'd factor these clues into the decision, so it wasn't purely a visual question of how well could the texture of a wall be discerned. But in the final print it was a purely objective, is that brick or just an orangey brown splodge? Yes or no.
-
I think the brain instantly filters out much of what you see so you can make very fast decisions in terms of motor skills, responding to the important things in your environment, etc..
When you look at a photo much of that is flipped off since it's a 2-D representation, that's when you notice all the little details, and that's where a sharp photo comes into it's own.
Lately I've been shooting with a tripod and stopping down more then usual. One thing I am definitely noticing is the more depth of field you have and the sharper the image, the more critical it is to notice all those details in the shot. Because you are going to notice them when you look at the print even though your brain filtered them out when you visualized the shot.
Relevant to the Essay about lifting the shroud I guess.
-
One thing that is in the future too is not better interpolation software, which is formidable and will get better, but something that will definitely be better.
I don't know how long it will take, but I do know that a bit map type image, like RAW files, converted will be HUMONGOUS, like maybe 100GB or more--someone else will need to do that math.
I'm talking about bmp to vector conversion. Once that happens, you can use any MPs necessary to get the detail you want, and you will lose nothing to infinity when enlarging. I remember when early versions of Flash came out it had a bmp to vector conversion option. It was buggy and would crash your computer half the time, or take 2 days to convert a 600 x 600 BMP to a vector. But it did work kinda. However, a 100K file turned into a 10 MB file--when it worked.
When we do get to true vector images, or true vector conversion, our enlargement troubles will be over forever. They are enlargeable to infinity with no loss of detail etc. It just takes a huge amount of stored calculations to reproduce bmp images, and that is why the file containing those mathematical calculations is so huge. At least that is how I understand it generally.
BTW--Just how good is the best enlargement software nowadays?
-
An afterthought on my previous comments: I expect that we will move to sensors having noticeably higher resolution than lenses, so that resolution will be noticeably lens limited, but with sensors still holding resolution a bit below the ultimate potential of the lenses. (Apparently this has been so with larger format for some time.)
I expect this because
- When a lens and sensor have roughly equal resolution, the combined resolution is noticeably less that that of either component, so simultaneously the lens holds back the resolution potential of the sensor and vice versa.
- Thus, increasing sensor resolution somewhat beyond that of the lens can still improve overall resolution, getting closer to squeezing the most out of the lens. (The reverse is true too: increasing lens resolution beyond that of the sensor will also improve overall resolution.)
- There is now far more potential for increasing sensor resolution than lens resolution, and the cost of increasing sensor resolution is far lower, so I expect sensors to move ahead of lenses rather than the other way around.
In other words, the component of a system that is the most difficult and expensive to improve tends to end up being the weak link in performance, and I suspect that a collection of high end lenses is now that "most expensive and most difficult to improve" part of a high end DSLR system.
-
In other words, the component of a system that is the most difficult and expensive to improve tends to end up being the weak link in performance, and I suspect that a collection of high end lenses is now that "most expensive and most difficult to improve" part of a high end DSLR system.
[a href=\"index.php?act=findpost&pid=58846\"][{POST_SNAPBACK}][/a]
Do I get a sense here that you are coming around to my point of view on this issue, BJL . Ie., it might be easier to build an MF lens that's fully diffraction limited at f16 than it is to build a 35mm lens that's fully diffraction limited at f8. Or to put it another way, an MF lens that can deliver 50 lp/mm at 50% MTF at f16 is a more realistic proposition than a 35mm lens that can deliver 100 lp/mm at 50% MTF at f8. (For all I know, there might already exist such MF lenses capable of good contrast at 50 lp/mm and f16.)
-
I agree wiht this line of thought. There are limitations to anything, analog ( physical) being a point in case. Digital, however, can get better in spite of physical limitations, as long as mathmatical limitations are not met. As I stated above, we can at this time convert a digital image, or a film image to digital, and then convert that to a vector graphic, which is a mathmatical computation of tonality. Now we can rezise it to infinity without degradation. The technology, however, is not nearly ready for the consumer. How far off is that? I don;t know becsaue I haven't kept up with it, but barring any huge leaps in storage, storage speeds, and processing power in teh next 5 years, I'd say it's decades off. That's why Flash limits itself to flat colors, and recently some "fake" shadowing, which is really several layers of flat colors that are made to look like a featherd shadow. Note I didn't say "feathered edge" because true vector graphics do not suffer from artifacting nor jaggies.
Do I get a sense here that you are coming around to my point of view on this issue, BJL . Ie., it might be easier to build an MF lens that's fully diffraction limited at f16 than it is to build a 35mm lens that's fully diffraction limited at f8. Or to put it another way, an MF lens that can deliver 50 lp/mm at 50% MTF at f16 is a more realistic proposition than a 35mm lens that can deliver 100 lp/mm at 50% MTF at f8. (For all I know, there might already exist such MF lenses capable of good contrast at 50 lp/mm and f16.)
[a href=\"index.php?act=findpost&pid=58871\"][{POST_SNAPBACK}][/a]
-
As I stated above, we can at this time convert a digital image, or a film image to digital, and then convert that to a vector graphic, which is a mathmatical computation of tonality. Now we can rezise it to infinity without degradation. The technology, however, is not nearly ready for the consumer. How far off is that? I don;t know becsaue I haven't kept up with it, but barring any huge leaps in storage, storage speeds, and processing power in teh next 5 years, I'd say it's decades off. [a href=\"index.php?act=findpost&pid=58880\"][{POST_SNAPBACK}][/a]
I think there's certainly room for improvement of interpolation algorithms and this will no doubt take place in step with increases in computer power, but I can't help getting the impression here that you have been unduly influenced by certain scenes in Hollywood movies . I think you might know the type I'm referring to. The CIA is given a satellite image of suspicious installations or camps in the Arabian desert but they can't quite make out what's going on, so they bring in the imaging guys with their state-of-the-art computers who miraculously zoom in on the minutest detail until the bearded face of Usama Bin Laden becomes as clear as a family album snapshot.
There's a fundamental flaw in such imagined processes. If the detail has not been recorded to begin with, then no amount of interpolation can accurately create it. For example, if your 3mp camera does not have the resolving power to capture say, a single strand of hair blowing in the wind when you take a portrait shot from a particular distance, then you can't expect an interpolation program to 'divine' that a single strand of hair in a particular location should have been recorded and 'create' it for you.
However, I would admit that a really smart computer might be able to recognise tell-tale signs of windy conditions in a scene as well as a human head of hair and recreate a few single strands for the sake of realism, but that would be of no use to the CIA. They might end up arresting the wrong person.
-
Do I get a sense here that you are coming around to my point of view on this issue, BJL. Ie., it might be easier to build an MF lens that's fully diffraction limited at f16 than it is to build a 35mm lens that's fully diffraction limited at f8.
[a href=\"index.php?act=findpost&pid=58871\"][{POST_SNAPBACK}][/a]
No.
Mostly because I was not comparing performance of different formats at all!
Also because I think that your standard of "maximum diffraction limited resolution" is a red herring, ignoring the real-world balancing of diffraction, aberration, out of focus effects, sensor resolution, as well as bulk and cost. What I was thinking of with my previous post was roughly "lines per picture height with appropriate DOF, within a given budget". And I believe that the most cost effective strategy for increasing the resolution of images is the one being almost universally pursued by sensor and camera makers: increasing pixel density at each sensor size, (while keeping sensor sizes the same or reducing them).
-
No.
Mostly because I was not comparing performance of different formats at all!
Also because I think that your standard of "maximum diffraction limited resolution" is a red herring, ignoring the real-world balancing of diffraction, aberration, out of focus effects, sensor resolution, as well as bulk and cost. What I was thinking of with my previous post was roughly "lines per picture height with appropriate DOF, within a given budget". And I believe that the most cost effective strategy for increasing the resolution of images is the one being almost universally pursued by sensor and camera makers: increasing pixel density at each sensor size, (while keeping sensor sizes the same or reducing them).
[a href=\"index.php?act=findpost&pid=58933\"][{POST_SNAPBACK}][/a]
BJL,
That large format cameras tend to be bulkier and heavier than small format cameras is a given. We can't do much about that, although it is perhaps worth mentioning that my Fuji GSW-690 lll, a 6x9cm film format camera with fixed wide-angle lens, is considerably lighter than a 1Ds with equivalent 35/1.4 lens and hardly heavier than a D60 or 5D with the same lens. Of course the GSW-690 lll has the weight advantage of being a manual rangefinder camera with fixed lens and carbon fibre body. It's designed to be light.
What I find interesting in your argument above is the way you've ignored technological solutions to the DoF disadvantage of the larger format whilst advocating other technological solutions to the dynamic range disadvantage of the smaller format.
I'm not qualified to comment on the technological difficulty of rapidly filling and emptying small photosites within one long exposure that is more comparable to the longer exposures of the larger formats with their bigger f stops, but the technology of merging a number of slightly differently focussed shots to produce one image with tremendous DoF, already exists. All that remains is for that process to be refined and automatized within the camera.
I also find it slightly frustrating that, despite your avowed knowledge in the field of optics and lens design, you seem unable to shed any light on this vexed question of precisely at what apertures are lenses diffraction limited. I hope you don't think I'm trying to 'put you on the spot' here. I mean, nobody else on these forums seems to be able to comment on these matters either. Michael seems to have lost interest in his DXO Analyzer, but the few lenses he has tested sometimes show surprising performance variation at big f stops where one might expect diffraction to dominate and be the great equaliser.
This might seem like a red herring, but I don't believe it is. Sensor prices are coming down for both large and small formats. If we ever get to the stage when all sensors will have smaller pixels than the smallest detail their associated lenses can resolve, there may well be an economic/quality shift to the larger format as the lens limitations of the smaller formats (ie 35mm) are reached.
-
No, I didn't say that. Vector conversion will not add information to the picture. However, what you capture will be resizable--blurry, fuzzy, or what have you--to infinity with no further degradation. I was influenced not by Hollywood but by the mathmatics behind vector creation, and then had my first practical experience--with Flash--by converting small graphics to vector images, and tehn resizing them to 200 time their original size--with no loss in quality compared to the original at the originals size. You never see dots or blocks--ever. The problem is that of today, vector graphics are used mainly for flat drawing type images, wiht teh exception of CADD systems and programs like Maya, which are still animation type drawings, and not a ceonversion of a true photograph in to a true vector image--the tonality of a real picture, like I said, would take days to convert to a true vector image and would be just HUGE in size. Even Maya rendering into a vector graphic on a dual Mac G-X whatever the top of the line Mac comuter is sometimes take more than 24 hours, or more. Mathmatical computation of tonality will eventually take over bit maps. There will need to be vast increases in computational power along with vast improvments in storage before this comes about. We're talking about crytal storage devices that act like solid state RAM, but vastly faster and larger--like as in a desktop computer handling RAW images and converting them would need to be something like a Petabye, which is something like a 1, 000 times greater than a Terabyte, which is 1, 000 MBs. waht will really be interesting is when cameras can offer vector images themselves, again, probably when I'm around 200 years old.
I think there's certainly room for improvement of interpolation algorithms and this will no doubt take place in step with increases in computer power, but I can't help getting the impression here that you have been unduly influenced by certain scenes in Hollywood movies . I think you might know the type I'm referring to. The CIA is given a satellite image of suspicious installations or camps in the Arabian desert but they can't quite make out what's going on, so they bring in the imaging guys with their state-of-the-art computers who miraculously zoom in on the minutest detail until the bearded face of Usama Bin Laden becomes as clear as a family album snapshot.
There's a fundamental flaw in such imagined processes. If the detail has not been recorded to begin with, then no amount of interpolation can accurately create it. For example, if your 3mp camera does not have the resolving power to capture say, a single strand of hair blowing in the wind when you take a portrait shot from a particular distance, then you can't expect an interpolation program to 'divine' that a single strand of hair in a particular location should have been recorded and 'create' it for you.
However, I would admit that a really smart computer might be able to recognise tell-tale signs of windy conditions in a scene as well as a human head of hair and recreate a few single strands for the sake of realism, but that would be of no use to the CIA. They might end up arresting the wrong person.
[a href=\"index.php?act=findpost&pid=58903\"][{POST_SNAPBACK}][/a]
-
As I stated above, we can at this time convert a digital image, or a film image to digital, and then convert that to a vector graphic, which is a mathmatical computation of tonality. Now we can rezise it to infinity without degradation.
Well, no, it just doesn't work like that. To get the benefits of vector graphics that you mention, you have to have an accurate mathematical description of the image. But in general, there isn't any such description of a photograph. To be more precise, you can get a mathematical description (e.g., a spatial fourier transform and a wavelet transform), and we can do that now (given lots of computing power), but it won't help blow the print up to arbitrary sizes.
That's because the basic information just isn't there in the data. You're limited by the pixel density. Otherwise, good interpolation algorithms would already let us blow up pictures way bigger they they can be. Basically, you can't know what happens between the pixels, so you have to assume that there is only a smooth variation between them. But at a very large, blown up scale, a real subject would have more detail in between the (now spaced out by enlargement) pixels. So the enlargement looks artificially smooth. And using vector graphics can do nothing about this.
-
So what type of enlargement can you get with vector graphics compared to bitmap, once the conversion is made.
Also, are you saying that we will never had the ability to define a shape with tones in a image mathmatically that will allow an infinite--not arbitrary--enlargement--which means keeping all of the tonal scales that are originally in the image?
Well, no, it just doesn't work like that. To get the benefits of vector graphics that you mention, you have to have an accurate mathematical description of the image. But in general, there isn't any such description of a photograph. To be more precise, you can get a mathematical description (e.g., a spatial fourier transform and a wavelet transform), and we can do that now (given lots of computing power), but it won't help blow the print up to arbitrary sizes.
That's because the basic information just isn't there in the data. You're limited by the pixel density. Otherwise, good interpolation algorithms would already let us blow up pictures way bigger they they can be. Basically, you can't know what happens between the pixels, so you have to assume that there is only a smooth variation between them. But at a very large, blown up scale, a real subject would have more detail in between the (now spaced out by enlargement) pixels. So the enlargement looks artificially smooth. And using vector graphics can do nothing about this.
[a href=\"index.php?act=findpost&pid=59663\"][{POST_SNAPBACK}][/a]
-
However, the quality test that I'd advocate is sterner; I'd ask at what point does an enlargement start to look significantly inferior to one made from a larger negative? In the days of film IMO a 35mm could generally hold its own up to a 10"x8" print, for a really good negative you could push that up to an 11"x14". After which the challenge becomes almost impossible as you pushed further into medium format territory.
What I see in my prints (using a 5D, 1Ds Mk II, P25 with a Hasselblad, and a P25 on a Linhof movement camera stitched to a sensor equivilant size of 49mmx71mm) is that even though digital gives a little more enlargement capability than film, especially with longer prime lenses, it's not that much more. And why should it be?
To go back to the 28MP Canon you mentioned (and I agree that something like it is sure to arrive before too long), some will look at that pixel spec and conclude that 240 dpi will get them pretty close to a 20" x 30" print. From 35mm? Surely it's time for a sanity check. What would the practical depth of field constraints be for an 18x or 20x enlargement? What back-breakingly heavy tripod would be required to still every last micron of vibration? And if there's a growing sense of dissatisfaction now with Canon's wide angles, what would the feeling be when those same optics are scrutinised at massive enlargement scales?
[a href=\"index.php?act=findpost&pid=58624\"][{POST_SNAPBACK}][/a]
Hi Gary, as others have suggested in this thread, I think the right way to look at the enlargement issue with digital is to look at the number of pixels. (Of course there's the issue of the quality of the pixels themselves, but let's put that issue aside for now.)
Now, the human eye can resolve features as small as ~1 minute of 1 arc, assuming 20/20 vision. Let's say the eye is 6 inches away from the print, i.e. we really want to examine features closely. At this distance, the average eye can resolve features as fine as 0.0017 inches, i.e. two black lines spaced any closer than that will start to appear as a single gray line with no space in between. This means that we need at least 285 pixels printed per (linear) inch. More than that will yield diminishing returns, even for people with unusually good vision. This number holds regardless of the size of the print itself.
So, to get back to your original question: if a digital system can produce 285 "good" pixels per linear inch, then the print should be rich with detail.
Example: the 1Ds Mark II has a native file resolution of 4992 x 3328. Without any interpolation, this yields a printed file roughly 11.5 x 17.5 in size @ 285 pixels per linear inch -- no problem. Of course, we can make larger prints through interpolation, but in doing so we can only add artificial detail.
Eric
-
What you would really need to build a digital camera that could create arbitrarily size enlargements is a device that does not capture a 2D projection of the 3D world at all.
What you need is something that can build a programmatic description of the entire scene and all the light sources involved. Then you can render the entire scene at an arbitrary size just like raycasting/tracing software.
If you get behind the scenes a little and ignore the big commercial applications with their simplified user interfaces this is what they're doing, they're building a programmatic description of a scene, when you render it, you are projecting it into a two dimensional image. This is the same way video games work, the renderings can be done in hardware or software.
There is no such device to capture a model of an arbitrary scene just by pointing a camera at it. If I was to envision a device that could do this it would have to be some kind of floating robot which flew around the area measuring everything with robots.
And even then everything I just mentioned ignores that even with the model of everything in the scene and all the lighting information you still need textures of every object in the scene at an arbitrary level of detail.
It's orders of magnitude more complex then converting a 2D bitmap to a 2D vector graphics description in flash.
There is a lot of good information about this in computer science texts.
One really fun place to play around with this is POVRay, the free ray tracer... lots of fun to learn to program that. I played around with it a lot about 10 years ago in college. There is a lot of fantastic art being done with it.
Just don't hold your breath for a device or computer program which can capture that kind of information from a photograph.... that's very much in the realm of science fiction today.
Check out the internet ray tracing competition too... lots of awesome stuff to look at.
Internet Ray Tracing Competition (http://www.irtc.org)
These could all be rendered at arbitrary sizes.. but you've got to remember they are the product of:
- Someone composing the scene mathematically
- Lots of 2 dimensional textures captured with a camera or produced in photoshop, etc..
- Lots and lots of computational power & time. (Hours, days even)
-
What I find interesting in your argument above is the way you've ignored technological solutions to the DoF disadvantage of the larger format ... merging a number of slightly differently focussed shots to produce one image with tremendous DoF, already exists.
you seem unable to shed any light on this vexed question of precisely at what apertures are lenses diffraction limited.
[a href=\"index.php?act=findpost&pid=58959\"][{POST_SNAPBACK}][/a]
Firstly, what DOF disadvantage of larger formats? Use the larger format with same aperture diameter (so higher aperture ratio) and you get the same DOF, and then using the same shutter speed (but higher ISO) will give similar noise levels (mayne a little higher), and you also have the option of using same ISO (so lower shutter speed) to get less noise and more dynamic range.
The option you describe of merging multiple frames for increased DOF is clearly a very "low speed" one, for working with completely stationary subjects and a tripd mounted camera. That is exactly the territory where I do expect larger formats to continue to have an advantage, with electronic sensors just as with film.
As to diffraction limitation, all I can say in general (as opposed to about particular current lenses) is that to avoid losing a significant amount of the resolution that is possible from a Bayer CFA sensor of a particular pixel spacing, it seems that the f-stop should not be higher than about twice the pixel spacing. So about f/8-f/11 for the smallest current DSLR pixels of 5.3-5.4 microns. Thes f-stops are easily high enough that good lenses are performing at very close to the diffraction limited performance of a "perfect" lens. Indeed, looking at some good recent lenses, I would not worry about lens abberations significantly limiting resolution until about f/4 - f/5.6, corresponding to about 2 - 2.8 micron minimum useful pixel spacing.That gives 20-40MP in 4/3 format, 30-60MP in EF-S/DX, etc.
I do not expect 4/3, EF-S or DX format to ever give print resolution significantly beyond what the current 22MP "digital large format" sensors offer. Then again, I do not expect more than one in a thouand SLR users to care about such extremes of resolution either ... especially if it involves fiddling with taking and merging multiple exposures to get adequate DOF.
-
How many people are actually merging multiple shots for DoF on a regular basis?
Sounds like way more of a PITA then it's worth. Why not go to a tilt/shift lens or a medium/large format camera with camera movements if you need something that specialized?
Merging the files has got to be even more of a PITA then using film and then having to scan it.
Merging for exposure sounds like a PITA to me, merging for DoF sounds even more crazy... talk about a post processing nightmare.
-
Example: the 1Ds Mark II has a native file resolution of 4992 x 3328. Without any interpolation, this yields a printed file roughly 11.5 x 17.5 in size @ 285 pixels per linear inch -- no problem.
Eric, I'm afraid we're going to have to agree to disagree on this point. In the earlier days of digital we were mainly limited by the number of pixels at our disposal. So it made sense to measure enlargement capabilities just in terms of pixels, using the 240-360 dpi measure that you referenced. However, we're now getting to the stage IMO where we have to temper that simple linear relationship with some other factors.
Let me give you one example, depth of field. The DOF engravings on manual focus lenses were established in the 1930's, apart from the ASA/ISO standards for film it's one of the few examples where the photographic industry has ever agreed on anything! Rather than have different lens and camera manufacturers competing on who's products gave the greatest depth of field they agreed on a convention that said the image recorded on 35mm film would be considered as "in focus" if an infinitesimal spot in the subject was rendered in the negative as a "blur circle", or "circle of confusion" no more than about 0.030mm in diameter. There was in fact some departure from this standard, Leica engraved their lenses to a slightly more rigorous depth of field standard of about 0.025mm. But basically the convention has held for the last 70 years.
And if you enlarge a 35mm negative say x4, to get the common 4"x6" print, then it's a pretty good guide. That will render the point in the subject as a little blur circle about 0.1mm in diameter. Not quite up to the standard you spoke about but pretty good. And in the good old days of silver darkrooms 35mm enthusiasts like myself regularly achieved really sharp and satisfying 10"x8" prints, and on rare occasions with a really lucky negative we could even go a bit larger.
But suddenly with 11MP, 12MP, and 17MP DSLR's we seem to think that we can enlarge to x12 or x15 with impunity. Well, that takes the blur circle on the print up to about half a millimetre, and even to my middle aged eyes that just looks plain soft! Let alone guaged against the stern standards you've advanced for the limits of resolution.
Let me give you a practical example. Here's a 100% crop from a P25 back with a Hasselblad 100mm lens, it's precisely focused at a distance of 10 metres at an aperture of f11, no sharpening, no nothing.
In the second shot I've shifted the focus by turning the 10m focus mark around the lens, not to the f11 position that Zeiss/Hasselblad say is acceptable, but just a little way to the f5.6 position. So two full stops more restrictive than the recommended DOF limits. To my eye it looks like rubbish!
This is why I say that we've now moved beyond pixel limited enlargement barriers, and we're basically back to the same physical and optical constraints that have capped our photographic ambitions for the last seventy years.
[attachment=316:attachment]
[attachment=317:attachment]
-
Let me give you one example, depth of field.
[a href=\"index.php?act=findpost&pid=59820\"][{POST_SNAPBACK}][/a]
Aha, Gary -- now I see what you are getting at. I misunderstood your original post.
What you are saying is that there are other factors besides the # of pixels that affect the enlargeability, with optical phenomena (or limitations, so to speak) such as depth of field playing a role. To put it another way, there are these other factors that can limit the quality of each pixel, and therefore the overall potential to magnify is diminished. I think we're in agreement here ...
Eric
-
As to diffraction limitation, all I can say in general (as opposed to about particular current lenses) is that to avoid losing a significant amount of the resolution that is possible from a Bayer CFA sensor of a particular pixel spacing, it seems that the f-stop should not be higher than about twice the pixel spacing.
BJL,
Not sure what you mean here. The problem for most of us when thinking about lens performance is that the results we are used to seeing have either been filtered through the crud and large grain-clumps of film and then re-photographed with a scanner, or they are simply limited by a combination of digital noise and insufficient pixel density of the sensor and/or insufficient sensitivity and quantum efficiency of the photosites.
Either way, we never get to appreciate what a camera lens is really capable of.
My understanding is that a lens that is resolution limited only by diffraction at f8 should be capable of double the resolution of a lens that is diffraction limited at f16.
The difference in resolution between these two f/stops should be comparable to the difference in total picture resolution between the D30 (not the 30D) and the 5D., ie. 4x the pixel count equates to double the resolution.
When I compare f16 with f8, using either my 5D or 20D, I do not see a substantial difference. I do see some difference, but such differences seem far less significant than the differences I sometimes see when comparing two similar lenses in a store before buying. I therefore can't help getting the impression that there really are no 35mm lenses currently available that are diffraction limited at f8.
Furthermore, this impression is reinforced by statements from many professional photographers who claim that 35mm lenses are now the weak point in the chain and futher pixel density will not serve any purpose without improvement in lenses.
If for argument sake, one of my lenses is diffraction limited at f16 and transmits detail at 50lp/mm to my sensor at 50% MTF, I would expect a result that is quite close to the resolution limits of a 5D sensor. If a 35mm lens exists which is diffraction limited at f8, then by the same standard it should be capable of transmitting detail equivalent to 100 lp/mm at 50% MTF.
To capture such detail on a Bayer type 35mm sensor, one would need a 48mp sensor (4x the pixel count of the 5D) with 'yet to be developed' technology to reduce noise and increase sensitivity. Would this not be the case? Seems logical to me .
-
But suddenly with 11MP, 12MP, and 17MP DSLR's we seem to think that we can enlarge to x12 or x15 with impunity. Well, that takes the blur circle on the print up to about half a millimetre, and even to my middle aged eyes that just looks plain soft! Let alone guaged against the stern standards you've advanced for the limits of resolution.
Gary,
If I've understood your post, you are actually giving an example here of something BJL and I have been discussing on other threads. Increased resolution pays the penalty of shallower DoF.
In all DoF calculators, a specified CoC is required (or assumed by the calculator). If you intend blowing up your digital images to their maximum size at 240ppi without interpolation (about 14"x21" with the 1Ds2) and if you insist on scrutinising these largish prints from the usual close distance you would view a postcard, then the CoC size for DoF purposes should be equal to the pixel pitch of your sensor.
If you are aiming for 'pixel sharpness', which probably requires detail at the resolution limits of the sensor to be transmitted by the lens at 'at least 50% MTF' and possibly 70%, then on a print at 240ppi without interpolation (14x21" with 1Ds2) anything that's not in the focal plane, ie. not in fous, will be seen as such, ie., slightly (or greatly) OOF.
This is actually a serious problem for all pixel peepers. It's due to the fact that a lens can only ever be truly focussed at one plane (although I suspect this might not be true in an absolute sense).
A shining example of this principle would be the pin-hole camera. It achieves tremendous DoF by not being sharp anywhere .
-
BJL,
Not sure what you mean here.
When I compare f16 with f8, using either my 5D or 20D, I do not see a substantial difference. ... I therefore can't help getting the impression that there really are no 35mm lenses currently available that are diffraction limited at f8.
[a href=\"index.php?act=findpost&pid=59830\"][{POST_SNAPBACK}][/a]
Ray,
which lenses are you talking about?
Do no presume that because your Canon lenses designed for 35mm film cameras (and targetted at good MTF at 10 and 30 lp/mm) are abberation limited rather than diffraction limited at f/8, then so are all current lenses, including ones specifically designed for smaller image circles and the higher resolution of current and future electronic sensors. The other authorities you refer to seem also to be talking about 35mm film camera lens designs.
On the other hand, Thom Hogan sees diffraction limitation with the D2X and Nikon (DX?) lenses from about f/11, and so definitely a difference between f/8 and f/16. And you might remember that Didger said a similar thing about combinations like the D2X and 12-24 DX, making f/8 his diffraction based limit. To examine the state of the art for lp/mm resolution in sensor/lens combinations, you might want to look at a combination like a D2X with a good Nikon prime, or an Olympus E-500 or E-330 with 50/2 or 35/3.5 macro.
-
Do not presume that because your Canon lenses designed for 35mm film cameras (and targetted at good MTF at 10 and 30 lp/mm) are abberation limited rather than diffraction limited at f/8, then so are all current lenses, including ones specifically designed for smaller image circles and the higher resolution of current and future electronic sensors. The other authorities you refer to seem also to be talking about 35mm film camera lens designs.
BJL,
My highest resolving lens is the Canon 50/1.8 ll with a quite respectable Photodo rating of 4.2. I remember the occasion when I bought it. I was curious if this lens could actually deliver more 'absolute' resolution than my medium quality Canon 28-135 IS zoom at 50mm, using my D60 and photographing a Norman Koren test chart. Surprisingly I found that it couldn't.
I then did a number of 'real world' field tests and assiduously searched for detail at 200% and 400% on my monitor, that might exist in the 50/1.8 shot (at f8) and not in the zoom shot at the same focal length. Surprisingly again, I could not find any such detail.
What I did find, however, was greater accutance in the 50/1.8 shots, and that gave the impression that all 50/1.8 shots were slightly sharper. Even the test chart lines appeared sharper, although there were no more of them to be seen.
For this reason, I use my 50/1.8 in place of the 28-125 IS zoom whenever the shot requires a 50mm FL, provided I am not in a situation where IS would be a definite advantage, and provided I am not feeling too lazy to change lenses.
However, this 'accutance' concept causes me to wonder just how much a really expert and experienced PS user familiar with the best of the 3rd party sharpening routines might be able to achieve in narrowing that accutance gap between the two lenses.
In general, I'd agree that it's better to start off with a 'natural' sharpening, but I'm still surprised at that difficulty I experienced in finding just one iota of significant detail that the higher resolving 50/1.8 might be able to deliver, with a camera, by the way, that has the same resolving power as the 1Ds2. (lp/mm not total picture resolution, for the benefit of those who might get confused).
-
BJL,
My highest resolving lens is the Canon 50/1.8 ll
[a href=\"index.php?act=findpost&pid=59950\"][{POST_SNAPBACK}][/a]
That is what I thought. Comparing MTF charts, the 50/1.8 II has MTF at 30lp/mm and f/8 comparable to what the Olympus 35/3.5 macro has wide open at f/3.5 and at twice the resolution, 60lp/mm. Except that the 35/3.5 holds resolution far better towards the corners of the frame. Some of the new Schneider and Rodenstock "digital large format" lenses are comparably good at 60lp/mm.
If this appearance that thes new lenses designed for digital cameras roughly double resolution is real, it would move the threshhold of diffraction limitation down to half the f-stop (so it is nice to be able to compare the 35/3.5 at roughly half the f-stop of the 50/1.8.) If so, your observation of diffraction limitation starting between f/8 to f/16 probably moves to the range f/4 to f/8 with at least some of the new digital oriented lenses.
-
That is what I thought. Comparing MTF charts, the 50/1.8 II has MTF at 30lp/mm and f/8 comparable to what the Olympus 35/3.5 macro has wide open at f/3.5 and at twice the resolution, 60lp/mm. Except that the 35/3.5 holds resolution far better towards the corners of the frame. Some of the new Schneider and Rodenstock "digital large format" lenses are comparably good at 60lp/mm.
When I first read the specs of the new Rodenstock 35mm f4 HR, I immediately thought, 'this lens looks as good as some of those Zuiko primes for the 4/3rds format, yet with a much larger image circle'. What's also surprising is that it's a retrofocus design. It would therefore seem that lens design and technology has reached the stage where it should be possible for Canon (perhaps through a contractual deal with Rodenstock) to start offering us Zuiko quality lenses.
Nevertheless, there is a conundrum in this 'sensor quality versus lens quality' issue which I'm having difficulty fathoming.
I've already established for my own purposes that a good prime like the 50/1.8 (good within the panoply of Canon lenses, that is) is capable of imparting greater accutance to an image than a lesser lens, but not capable of greater 'absolute' resolving power of any significance. Of course, the lens is almost certainly capable of greater absolute resolving power than the 28-135 IS at 50mm, but the sensor is capping the results.
This debate about current lenses being inadequate for current sensors is really all about accutance rather than absolute resolution. Just as my modest 50/1.8 produces images that 'appear' sharper than an equivalent image from a zoom lens without providing additional detail in any obvious way, a Zuiko quality lens or Rodenstock HR lens on one of my Canon DSLRs would, I expect. produce images with even greater accutance but probably only a very marginal increase in absolute resolution.
Okay! Here's the conundrum, in big letters in case it's missed.
WHY DO THE OLYMPUS 8MP 4/3RDS CAMERAS NOT SHOW THIS GREATER ACCUTANCE WHICH ONE WOULD EXPECT FROM THE SUPERIOR ZUIKO LENSES?
(compared with the 20D and its low MTF Canon lenses)
-
How many people are actually merging multiple shots for DoF on a regular basis?
Sounds like way more of a PITA then it's worth. Why not go to a tilt/shift lens or a medium/large format camera with camera movements if you need something that specialized?
Merging the files has got to be even more of a PITA then using film and then having to scan it.
Merging for exposure sounds like a PITA to me, merging for DoF sounds even more crazy... talk about a post processing nightmare.
[a href=\"index.php?act=findpost&pid=59802\"][{POST_SNAPBACK}][/a]
Actually more people than you might imagine, and it's extremely easy to do - not a PITA at all. The issue is that for some situations it's the "only" solution where for others it's not the best solution.
Let's take a few examples. Let's say you are doing photomicrography of cell structures or even something as large as tiny insects. A microscope has tremendously small DOF so it's impossible to get all the subject into focus simultaneously. With easily obtainable software I can take a number of frames each with separate focus points using the "best" optical sharpness aperture for the particular lens and combine them into an infinite focus single image in seconds.
To carry this further and even into the realm of landscape photography. I can shoot multiple frames (even as few as three) using the sharpest aperture of my lens and combine them into a single hyperfocal landscape image which is "better" than anything I can get with any single lens.
Now the down side. If the subject or something else moves between frames due to wind, intrusions into the frame from animals, birds, insects, etc., then this method fails. The other downside is that it take two or more frames to accomplish this. But don't sell this short, it's a viable means of increasing DOF and producing incredible results which can't be accomplished any other way.
Lin
-
With easily obtainable software I can take a number of frames each with separate focus points using the "best" optical sharpness aperture for the particular lens and combine them into an infinite focus
That sounds interesting, can you tell us more about this software?
By the way, I stitch shots all the time with a Phase One P25 on a large format camera, there's several manufacturers now making "sliding carriage" backs for large format cameras. You basically attach the digital back to the carriage and shuffle it around inside the image circle, you don't need software to bring the frames together, the sliding carriage handles it all with pixel accuracy.
I use it for architectural work where it's a breeze. Even if there's people or cars in the shot unless they're actually straddling the dividing line it doesn't matter, even then you can Photoshop them out. The only really tricky problem is when photographing a building across water, then stitching becomes a marathon Photoshop exercise in blending in waves and ripples. But for a 20" x 30" print this delivers the goods with sumptious, "nose to the print" quality.
-
WHY DO THE OLYMPUS 8MP 4/3RDS CAMERAS NOT SHOW THIS GREATER ACCUTANCE WHICH ONE WOULD EXPECT FROM THE SUPERIOR ZUIKO LENSES?
[a href=\"index.php?act=findpost&pid=60043\"][{POST_SNAPBACK}][/a]
What evidence of accutance comparisons are you looking at? The only resolution comparisons I have seen are at the opposite extreme from accutance/contrast, being comparisons of "extinction resolution". What is more, these are based on comparisons of standard JPEG output from different cameras. This means that sensor performance, low pass [AA] filter performance, Bayer interpolation algorithms, and JPEG conversion algorithms are all affecting the results, obfuscating attempts at pure lens to lens comparisons. (Comparing image quality with results from different cameras, one far older and/or cheaper than the other, would be a rather unscientific way of comparing lens performance.)
-
That sounds interesting, can you tell us more about this software?
By the way, I stitch shots all the time with a Phase One P25 on a large format camera, there's several manufacturers now making "sliding carriage" backs for large format cameras. You basically attach the digital back to the carriage and shuffle it around inside the image circle, you don't need software to bring the frames together, the sliding carriage handles it all with pixel accuracy.
I use it for architectural work where it's a breeze. Even if there's people or cars in the shot unless they're actually straddling the dividing line it doesn't matter, even then you can Photoshop them out. The only really tricky problem is when photographing a building across water, then stitching becomes a marathon Photoshop exercise in blending in waves and ripples. But for a 20" x 30" print this delivers the goods with sumptious, "nose to the print" quality.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=60234\")
Hi Gary,
The software was originally designed for photomicroscopy and there are several versions available from extremely expensive to free. The one's I'm familiar with are listed below with links. Of these three I use Helicon Focus (moderately priced) and Combine Z (freeware - donations accepted).
First, the very expensive but very powerful Auto Montage by Synchroscopy. This software does it all, but is out of the range for most except for the university academic or scientist who has a large budget for equipment and software.
[a href=\"http://www.syncroscopy.com/syncroscopy/am.asp]http://www.syncroscopy.com/syncroscopy/am.asp[/url]
Next is Helicon Focus - very powerful, almost as powerful as Auto Montage if you buy the "Pro" version. It even allows for a dust map to automatically remove (clone out) sensor dust. I use this one most....
http://heliconfilter.com/pages/index.php?id=509 (http://heliconfilter.com/pages/index.php?id=509)
Finally CombineZ which is also quite powerful but a little more difficult to use.
http://www.hadleyweb.pwp.blueyonder.co.uk/CZ5/combinez5.htm (http://www.hadleyweb.pwp.blueyonder.co.uk/CZ5/combinez5.htm)
Any of these products will do the job of combining same frame with differential focus points into an "infinite" focus single image. The differences are in how well they can handle "registration" (the slight shift in size of the image when choosing differential focus points). There are a few "caveats" you need to keep in mind when using these for hyperfocal landscapes - things like not having a huge tree six inches from the lens with mountain wildflowers three feet in front - etc. You will quickly get a feel for the possible, probable and impossible when doing this. If you are interested I think I still have a sample somewhere which I posted a year or so ago showing a digicam on a tripod about a foot from the lens with focus continuous from this spot to mountains several miles distant. Let me know and I'll dig it up. It shows the pitfalls and promise of this technique.
addendum:
Actually I just looked and found a link where I posted this test on the Helicon Focus forum. Here's the link if you want to see the original and combined frame.
http://helicon.com.ua/forum/viewtopic.php?t=121 (http://helicon.com.ua/forum/viewtopic.php?t=121)
Stitching:
Yes, stitching is definitely the poor man's BetterLight scanning camera. It offers a way to dramatically increase resolution while maintaining perspectives which are impossible to get with the same system and any wide angle lens. It's always great to see a stitched wide angle or panorama without nearby trees leaning in at an obtuse angle which would be the result of using a normal wide angle lens on a 35mm platform. I often look at excellent landscapes such as those done by the late Galen Rowell where this issue detracts from an otherwise incredible image. I suspect if Galen and Barbara were still alive today they would be shooting with a Canon 5D (he always loved lightweight camera bodies) or even a Nikon D2X and stitching to get some of those fantastic images he used to get with the little Nikon 35mm film bodies.
Best regards,
Lin
-
What evidence of accutance comparisons are you looking at?
BJL,
I'm looking at the standard studio shots at dpreview here (http://www.dpreview.com/reviews/olympuse500/page20.asp)
You'll notice in these shots of 'real' items, the Olympus E-500 has a slight advantage because the wider-aspect-ratio EOS 350D has been matched with respect to picture height. Despite this slight disadvantage of the 350D, it manages to deliver 'crisper' images than the E-500.
Both these cameras have the same pixel count (approx). The Olympus has the advantage of a better lens as well as the advantage of slightly more pixels in the vertical dimension, yet the 350D produces slightly better images with apparently greater accutance. This is particularly noticeable in the blow-up of the Leica lens featured in the studio shot.
What is more, these are based on comparisons of standard JPEG output from different cameras.
It's true that having only jpeg images to compare tends to obscure the results somewhat, if one wishes to put a really fine point on the matter. However, I've never heard the argument that a reason for shooting in RAW mode rather than jpeg is that all one's fine prime lenses will be rendered as mediocre zooms in jpeg mode.
Of course, it could well be the case that the finest quality jpegs the E-500 can produce simply do not do justice to the camera and that Canon have simply done a better job with regard to the 'fine quality' jpeg output of the 350D. If this is indeed the case, then point me to the evidence.
I can draw conclusions only from the evidence in front of me.
-
Lin, I've ordered the Helicon product, thanks for the tip.
-
Eric, I'm afraid we're going to have to agree to disagree on this point. In the earlier days of digital we were mainly limited by the number of pixels at our disposal. So it made sense to measure enlargement capabilities just in terms of pixels, using the 240-360 dpi measure that you referenced. However, we're now getting to the stage IMO where we have to temper that simple linear relationship with some other factors.
Let me give you one example, depth of field. The DOF engravings on manual focus lenses were established in the 1930's, apart from the ASA/ISO standards for film it's one of the few examples where the photographic industry has ever agreed on anything! Rather than have different lens and camera manufacturers competing on who's products gave the greatest depth of field they agreed on a convention that said the image recorded on 35mm film would be considered as "in focus" if an infinitesimal spot in the subject was rendered in the negative as a "blur circle", or "circle of confusion" no more than about 0.030mm in diameter. There was in fact some departure from this standard, Leica engraved their lenses to a slightly more rigorous depth of field standard of about 0.025mm. But basically the convention has held for the last 70 years.
And if you enlarge a 35mm negative say x4, to get the common 4"x6" print, then it's a pretty good guide. That will render the point in the subject as a little blur circle about 0.1mm in diameter. Not quite up to the standard you spoke about but pretty good. And in the good old days of silver darkrooms 35mm enthusiasts like myself regularly achieved really sharp and satisfying 10"x8" prints, and on rare occasions with a really lucky negative we could even go a bit larger.
[/But suddenly with 11MP, 12MP, and 17MP DSLR's we seem to think that we can enlarge to x12 or x15 with impunity. Well, that takes the blur circle on the print up to about half a millimetre, and even to my middle aged eyes that just looks plain soft! Let alone guaged against the stern standards you've advanced for the limits of resolution].
Let me give you a practical example. Here's a 100% crop from a P25 back with a Hasselblad 100mm lens, it's precisely focused at a distance of 10 metres at an aperture of f11, no sharpening, no nothing.
In the second shot I've shifted the focus by turning the 10m focus mark around the lens, not to the f11 position that Zeiss/Hasselblad say is acceptable, but just a little way to the f5.6 position. So two full stops more restrictive than the recommended DOF limits. To my eye it looks like rubbish!
This is why I say that we've now moved beyond pixel limited enlargement barriers, and we're basically back to the same physical and optical constraints that have capped our photographic ambitions for the last seventy years.
[attachment=316:attachment]
[attachment=317:attachment]
[a href=\"index.php?act=findpost&pid=59820\"][{POST_SNAPBACK}][/a]
All I know is that I have a 24x36 enlargement of a Ukrainian hotel I took at night with my 1DS Mk II , using a Canon 135 f/2and can see a tie on a person standing in the window of the hotel 3/8 of a mile away. I have some enlargements made with my 50 1.4 that blow me away.
drgary
-
...with my 1DS Mk II , using a Canon 135 f/2and can see a tie on a person standing in the window of the hotel 3/8 of a mile away.
drgary
[a href=\"index.php?act=findpost&pid=60321\"][{POST_SNAPBACK}][/a]
It's a pity you didn't have a 5D with you at the time. I'd love to see how that tie would look in a 5D shot .
-
BJL,
I'm looking at the standard studio shots at dpreview
...
I can draw conclusions only from the evidence in front of me.
[a href=\"index.php?act=findpost&pid=60258\"][{POST_SNAPBACK}][/a]
With that evidence from in-camera JPEGs at default settings, you can draw no conclusions at all: famously, different approaches to level of in-camera sharpening, AA filters and such make these images useless for comparing the resolution of lenses, or even or of sensors, at least when the results are vaguely close.
For example, my E-1 is permanently set at +3 sharpening (in a range of -3 to +5) for JPEG output; the default is quite soft.
I, as the 4/3 defence, feel no need to point you to the evidence; the "prosecution" has offered fataly flawed evidence and so not made its case. Why are you so determined to talk down a system and lenses you have never used?
-
"Do some relatively large, sharp photos reveal more detail than a person would discern when looking at the actual scene?"
No question about it. I've recently started using a 4x5 view camera (Cambo Legend) with a Better Light Super 6k scan back, mostly at 8000x6000 resolution. I took some pictures from my bedroom window when setting it up for the first time, and something that looks like a 2 pixel white splodge when looking full frame is actually a completely readable road sign at 100%, and definitely not readable by the naked eye (and I have extremely good eyesight due to laser eye surgery). So yes.
What I do find though is that at sensible print sizes, the Better Light images have more resolution than my eyes can resolve in the print, too. This has considerable visual impact with some images. (understatement!)