Estimating extra reach (resolving power) of crop vs FF

[ecka]

Here is a way of calculating the effective extra reach or resolving power of a crop body versus FF, which will amuse the geeks among us...

MTF is not a measure of resolution, it is a measure of a lenses ability to transmit contrast of the original scene through a lens. That's all it measures.

If what you really want is true image resolution, then there are two correct ways of measuring it. The first is a measurement of the performance of the lens itself. This measurement can be found in using something like a USAF resolution test chart and a microscope to perform aerial inspections of the image of the chart after it has passed through the lens.

Problems with this approach include difficultly in setting up a test bench (it's not easy at all). Additionally, the final result will not include measurements of chromatic aberrations, image distortions, field curvature, etc. The most important missing element in this kind of test is the exclusion of the imaging system itself. However, if the question is about real optical resolution, this test will give you the right answer.

Which leads to the second way of measuring, well, actually calculating real image resolution. This is diagnostic and very simple to perform. Simply take the number of image points ("pixels") in your file, divide by the size of your sensor (in millimeters), and divide by two. This number will represent the number of Line Pair per Millimeter (the measure of the ability to go from one white line to one black line) that your sensor can resolve.

This is rather interesting in that you can quickly see that a Canon 7D 18 mpixel sensor is capable of resolving 116 line pair per mm. The Canon 5D MkII is capable of 78 lppmm. As a comparison, Sony's 36mpixel FF sensor is capable of resolving 102 lppmm and Phase One's monster 80mpixel IQ180 returns 97lppmm.

Taking this a step further, look carefully at the physical limits (as in optical physics) of optical resolution, as measured in lppmm. You see that at f/2, an optically correct lens will return 695lppmm in the center of the scene where the light's wavelength is 589.3mu (green). At f/11, an optically correct lens will return 123lppmm, dropping off to 92lppmm at 25 degrees off-axis tangential.

Looking at this over the years, I have come to realize there is seldom a lens in-capable of resolving so poorly that a sensor (or old film for that matter) could out-resolve the lens. Sure, there are other optical effects, but we are talking pure resolution here. Nothing more.

Think about this for a moment. MTF does _not_ measure optical resolution. While useful, it does _not_ tell the story of resolution, no matter how much "math" you throw at it. Secondly, and perhaps most interestingly, optical physics show diffraction limited resolution at f/11 EXCEEDS currently manufactured sensors ability to return that resolution in all cases.

Rather shocking, don't you think?

Back to the original poster's point: Canon's 7D sensor outresolves (using the correct application of the word "resolution") the FF sensors from any manufacturer. It does so, however, for reasons other than those that were brought up.

Correct application of rational thought and real world science can help us properly understand and identify the errors and misleading comments widely published by marketeers and critics of optical imaging systems.

You see, if Canon had made a FF camera with 7D's pixel density, then there wouldn't be any questions - FF wins, period.
Nikon did that with their D7000/D7100 vs D800/D810. You crop FF and you get almost exactly what the pre-cropped D7000/D7100 produce. In fact, you can just shoot in DX mode on D800...
So there is no global question about crop winning anything vs FF other than price.
The question is why Canon didn't make that 46mp FF camera?
Was it because consumers didn't ask for it? - Maybe.
Why didn't they ask for it? - Because they are too ignorant in How-It-Works department. They are affected by this ... More_Pixels=More_noise=Bigger_Files=My_Old_iBox_Can't_Handle_it=Crop_Is_Just_As_Good_It's_Just_Magically_Different_Because_Size_Doesn't_Matter=I_Better_Buy_An_Overkill_Lens_For_My_Crop_And_Not_Use_The_Rest_60%_Of_It=I_Don't_Need_That_Much_Pixels_But_Digital_Zoom_Is_Bad#!@ERROR*2&$ERROR...System_Reset~Hello_Micro_Four_Thirds=Must_Buy_35-100/2.8_Because_It_Is_70-200/2.8 ... virus .

 

[ChristopherMarkPerez]

Here is a way of calculating the effective extra reach or resolving power of a crop body versus FF, which will amuse the geeks among us...

Which leads to the second way of measuring, well, actually calculating real image resolution. This is diagnostic and very simple to perform. Simply take the number of image points ("pixels") in your file, divide by the size of your sensor (in millimeters), and divide by two. This number will represent the number of Line Pair per Millimeter (the measure of the ability to go from one white line to one black line) that your sensor can resolve.

what about the bayer pattern?

a pixel in the image file is interpolated from photosites on the sensor.
so a bayer pattern sensor has to be worse than what you wrote?

You use the image dimensions from the file, not the actual pixels of the sensor. This is why I put "pixels" in quotes. I could have more accurately written "image information node in the output file" or something similar.

The point is, we want the total integration of the information at each "node" of information in the image file, since that's where the real world resolution will live as we process our images.

The calculation gives us the correct answer.

You could also take the slightly harder step of including the four pixels that make up an information node on the sensor itself, but the answer would still be the same in terms of calculating sensor resolution.

 

[Loswr]

The calculation gives us the correct answer.

How does your calculation account for the blurring introduced by the OLPF, or are you suggesting that doesn't affect resolution?

 

[ChristopherMarkPerez]

The calculation gives us the correct answer.

How does your calculation account for the blurring introduced by the OLPF, or are you suggesting that doesn't affect resolution?

When you look at the output of your image into a file, in terms of resolution testing, all that matters is that you can go from one file information node to another and transition from pure white to pure black (since that is how the USAF Resolution Test Chart-style tests work).

Indeed, you can observe this yourself by taking a 6:1 contrast ratio scene and look at the white to black transitions at 100 percent on your display. To avoid optical effects, I'd suggest stopping your taking lens to f/5.6 or f/8.

The effect is clear. If your sensor can make the transition from pure white to pure black between two information nodes in your file, your sensor/imaging/electronics system will give you pure resolution per the calculation.

I suspect that manufacturers are quite capable of managing this as every sensor I've ever had in my DSLRs (since the 40D) have behaved as hoped/desired. This is supported by the observations of every lens I've "tested" when attached to camera body.

I kept seeing the same numbers for USAF-style resolution _from wide open_ all the way down to f/11 and f/16 (for the earlier sensors). The only way to explain it was that the sensor is the limiting factor in terms of pure resolution (again, ignoring other optical effects).

 

[ChristopherMarkPerez]

You see, if Canon had made a FF camera with 7D's pixel density, then there wouldn't be any questions - FF wins, period.
Nikon did that with their D7000/D7100 vs D800/D810. You crop FF and you get almost exactly what the pre-cropped D7000/D7100 produce. In fact, you can just shoot in DX mode on D800...
So there is no global question about crop winning anything vs FF other than price.
The question is why Canon didn't make that 46mp FF camera?
Was it because consumers didn't ask for it? - Maybe.
Why didn't they ask for it? - Because they are too ignorant in How-It-Works department. They are affected by this ... More_Pixels=More_noise=Bigger_Files=My_Old_iBox_Can't_Handle_it=Crop_Is_Just_As_Good_It's_Just_Magically_Different_Because_Size_Doesn't_Matter=I_Better_Buy_An_Overkill_Lens_For_My_Crop_And_Not_Use_The_Rest_60%_Of_It=I_Don't_Need_That_Much_Pixels_But_Digital_Zoom_Is_Bad#!@ERROR*2&$ERROR...System_Reset~Hello_Micro_Four_Thirds=Must_Buy_35-100/2.8_Because_It_Is_70-200/2.8 ... virus .

I think it's more than that, really.

Think about this from the manufacturer's point of view. The electronics you use to integrate your system cost real money. The Bill Of Materials (BOM) will add up quickly as you select higher performing parts. Over time, the cost of parts will drop (typically) as newer, faster parts become available.

So it's a balance of price and performance.

From a business perspective, how much Gross Margin do you think you can recover? Answer that with the number of units sold and you'll be well on your way to finding out what the company's quarterly profit will be.

From an electronics point of view, proprietary data-busses (how data moves around inside your camera) are not cheap. Even when you build a system based on existing, open specifications, you still have to source the parts and make sure they will be available over the life of your product (not easy in this day and age). The more mega-pixels you design into your system, the faster your busses need to be to keep the beast from being Dog Slow.

I'm not saying it's not possible. Obviously, if Sony can say they'll release a 50mpixel FF sensor in a couple years, Canon could certainly do it too. But I think the companies are waiting for the BOM costs to drop, mixed with the corporate Gross Margin targets, into the region people will be willing to pay for a new toy.

 

[ecka]

You see, if Canon had made a FF camera with 7D's pixel density, then there wouldn't be any questions - FF wins, period.
Nikon did that with their D7000/D7100 vs D800/D810. You crop FF and you get almost exactly what the pre-cropped D7000/D7100 produce. In fact, you can just shoot in DX mode on D800...
So there is no global question about crop winning anything vs FF other than price.
The question is why Canon didn't make that 46mp FF camera?
Was it because consumers didn't ask for it? - Maybe.
Why didn't they ask for it? - Because they are too ignorant in How-It-Works department. They are affected by this ... More_Pixels=More_noise=Bigger_Files=My_Old_iBox_Can't_Handle_it=Crop_Is_Just_As_Good_It's_Just_Magically_Different_Because_Size_Doesn't_Matter=I_Better_Buy_An_Overkill_Lens_For_My_Crop_And_Not_Use_The_Rest_60%_Of_It=I_Don't_Need_That_Much_Pixels_But_Digital_Zoom_Is_Bad#!@ERROR*2&$ERROR...System_Reset~Hello_Micro_Four_Thirds=Must_Buy_35-100/2.8_Because_It_Is_70-200/2.8 ... virus .

I think it's more than that, really.

Think about this from the manufacturer's point of view. The electronics you use to integrate your system cost real money. The Bill Of Materials (BOM) will add up quickly as you select higher performing parts. Over time, the cost of parts will drop (typically) as newer, faster parts become available.

So it's a balance of price and performance.

From a business perspective, how much Gross Margin do you think you can recover? Answer that with the number of units sold and you'll be well on your way to finding out what the company's quarterly profit will be.

From an electronics point of view, proprietary data-busses (how data moves around inside your camera) are not cheap. Even when you build a system based on existing, open specifications, you still have to source the parts and make sure they will be available over the life of your product (not easy in this day and age). The more mega-pixels you design into your system, the faster your busses need to be to keep the beast from being Dog Slow.

I'm not saying it's not possible. Obviously, if Sony can say they'll release a 50mpixel FF sensor in a couple years, Canon could certainly do it too. But I think the companies are waiting for the BOM costs to drop, mixed with the corporate Gross Margin targets, into the region people will be willing to pay for a new toy.

Isn't 5D2 the most popular FF DSLR, which was an overpriced MEGApixel monster back in 2008 that forced everyone to buy larger capacity memory cards and new PCs; enjoy FullHD videos they didn't want and struggle shooting sports/action with it's inferior AF system and burst rate because they wanted those beautiful FF images?
All that whining about how much work has to be done and all the investments and new technologies ... is just unreasonable. We pay for it. We vote with our money for what we want. There may be 5D Mark 4 coming at Photokina and we can't stop it and I hope it will have 50+ mega pixels in it and 4K and maybe a hybrid viewfinder (OVF+EVF) and a 4" touch screen and quad-pixel AF or something and all that for only $3999 .

 

[Jack Douglas]

Phew, I can barely follow all the technical stuff in these CR threads but what seems to be clear to me is that the economics and market determine an awful lot more than we tend to realize and it's not all about who has the superior technology. Perhaps not unlike a board game with players carefully choosing their moves for fear they will find themselves trapped!

Jack

 

[Lee Jay]

Here is a way of calculating the effective extra reach or resolving power of a crop body versus FF, which will amuse the geeks among us.

Measure the MTF of a lens on the crop (= MTFcrop) and the same lens on the FF (= MTFff). The ratio of the MTFs, MTFcrop/MTFff, gives the relative resolving power of the bodies with that lens. However, the crop body can be placed 1.6x further away to give the same field of view. Therefore, the true effective relative resolving power, R, is given by:

R = 1.6x MTFcrop/MTFff.

Photozone lists measured MTFs for a set of lenses on the 5DII and 50D. I calculated their ratios for the Canon 200mm f/2.8 II, 85mm f/1.2 II and 35mm f/2 at wide apertures below the DLA. MTFcrop/MTFff is very close to 0.726 in all cases.

This gives R for 50D/5DII = 1.16.

So the effective extra reach is 16%. (Based on the ratio of their pixel sizes, a value of 36% is expected.

The dpreview widget gives values for the 5DIII and 7D only for a few lenses. I did the same calculations with the Tamron 150-600mm (between 150-400mm), the Canon 200-400mm and the Sigma 50mm f/1.4 A at wider apertures below the DLA. In all cases, MTFcrop/MTFff is close to 0.742.

This gives R for 7D/5DIII = 1.19.

So, the effective extra reach is 19%. (Based on the ratio of their pixel sizes, a value of 45% is expected).

There are always arguments about using MTFs quantitatively, but I think in this particular calculation it is reasonably valid to use them. It fits in reasonably well with experience - Jon has shown there is better resolving power in photos of the moon with the 7D, but it doesn't look 45% better. And my own experience is that the 7D and 70D aren't much better than the 5DIII, certainly not 1.6x.

There are numerous problems with this approach. You're assuming that MTF and resolution are both linear, when neither one is. You're also assuming MTF50 (which is what DPReview reports) is the correct measurement. It's not. Arguments rage, but many people use MTF 9, 5 or even 0 for "resolving power". Third, you're assuming DPReview's measurements are correct when in reality they aren't even self-consistent. I'll provide an example of that last one.

DPReview's measurements using Imatest: http://www.dpreview.com/lensreviews/widget/Fullscreen.ashx?reviews=84&fullscreen=true&av=4&fl=200&vis=VisualiserSharpnessMTF&stack=horizontal&&config=/lensreviews/widget/LensReviewConfiguration.xml%3F4

DxO's measurements using their own tool: http://www.dpreview.com/reviews/lens-compare-fullscreen?compare=false&lensId=canon_70-200_2p8_is_ii&cameraId=canon_eos5dmkii&version=0&fl=200&av=4&view=mtf-ca

At the same setting (200mm and f/4 on 7D), DPReview reports about 1500lp/ph while DxO reports about 1850lp/ph. Perhaps even more interestingly, DPReview reports about 1600 for the 5DII while DxO reports about 2400 for the 5DIII. So, the ratios are 1.6*1500/1600 (1.5) from DPReview for 5DII versus 7D or 1.6*1850/2400 (1.2) from DxO for the 5DIII versus 7D.

So which is it? The differences are very dramatic (1.5 versus 1.2) for cameras with almost exactly the same pixel size.

The facts are these: The actual ratio is always between 1 and the pixel density (PD) ratio, and where it lies in that range is a function of the quality of the technique and of the optics. The better the technique and the better the optics, the higher the number. With the better lenses used well, the actual number is much closer to the PD ratio than it is to 1.

Think of it another way - the PD ratio is like an optically-perfect teleconverter with magnification equal to the PD ratio. Now, real teleconverters aren't optically perfect, but the newer ones are pretty close.

Do teleconverters work when used on full-frame cameras? On the better lenses, yes they do. And so does the PD ratio.

 

[jrista]

Here is a way of calculating the effective extra reach or resolving power of a crop body versus FF, which will amuse the geeks among us.

Measure the MTF of a lens on the crop (= MTFcrop) and the same lens on the FF (= MTFff). The ratio of the MTFs, MTFcrop/MTFff, gives the relative resolving power of the bodies with that lens. However, the crop body can be placed 1.6x further away to give the same field of view. Therefore, the true effective relative resolving power, R, is given by:

R = 1.6x MTFcrop/MTFff.

Photozone lists measured MTFs for a set of lenses on the 5DII and 50D. I calculated their ratios for the Canon 200mm f/2.8 II, 85mm f/1.2 II and 35mm f/2 at wide apertures below the DLA. MTFcrop/MTFff is very close to 0.726 in all cases.

This gives R for 50D/5DII = 1.16.

So the effective extra reach is 16%. (Based on the ratio of their pixel sizes, a value of 36% is expected.

The dpreview widget gives values for the 5DIII and 7D only for a few lenses. I did the same calculations with the Tamron 150-600mm (between 150-400mm), the Canon 200-400mm and the Sigma 50mm f/1.4 A at wider apertures below the DLA. In all cases, MTFcrop/MTFff is close to 0.742.

This gives R for 7D/5DIII = 1.19.

So, the effective extra reach is 19%. (Based on the ratio of their pixel sizes, a value of 45% is expected).

There are always arguments about using MTFs quantitatively, but I think in this particular calculation it is reasonably valid to use them. It fits in reasonably well with experience - Jon has shown there is better resolving power in photos of the moon with the 7D, but it doesn't look 45% better. And my own experience is that the 7D and 70D aren't much better than the 5DIII, certainly not 1.6x.

There are numerous problems with this approach. You're assuming that MTF and resolution are both linear, when neither one is. You're also assuming MTF50 (which is what DPReview reports) is the correct measurement. It's not. Arguments rage, but many people use MTF 9, 5 or even 0 for "resolving power". Third, you're assuming DPReview's measurements are correct when in reality they aren't even self-consistent. I'll provide an example of that last one.

DPReview's measurements using Imatest: http://www.dpreview.com/lensreviews/widget/Fullscreen.ashx?reviews=84&fullscreen=true&av=4&fl=200&vis=VisualiserSharpnessMTF&stack=horizontal&&config=/lensreviews/widget/LensReviewConfiguration.xml%3F4

DxO's measurements using their own tool: http://www.dpreview.com/reviews/lens-compare-fullscreen?compare=false&lensId=canon_70-200_2p8_is_ii&cameraId=canon_eos5dmkii&version=0&fl=200&av=4&view=mtf-ca

At the same setting (200mm and f/4 on 7D), DPReview reports about 1500lp/ph while DxO reports about 1850lp/ph. Perhaps even more interestingly, DPReview reports about 1600 for the 5DII while DxO reports about 2400 for the 5DIII. So, the ratios are 1.6*1500/1600 (1.5) from DPReview for 5DII versus 7D or 1.6*1850/2400 (1.2) from DxO for the 5DIII versus 7D.

So which is it? The differences are very dramatic (1.5 versus 1.2) for cameras with almost exactly the same pixel size.

The facts are these: The actual ratio is always between 1 and the pixel density (PD) ratio, and where it lies in that range is a function of the quality of the technique and of the optics. The better the technique and the better the optics, the higher the number. With the better lenses used well, the actual number is much closer to the PD ratio than it is to 1.

Think of it another way - the PD ratio is like an optically-perfect teleconverter with magnification equal to the PD ratio. Now, real teleconverters aren't optically perfect, but the newer ones are pretty close.

Do teleconverters work when used on full-frame cameras? On the better lenses, yes they do. And so does the PD ratio.

Actually, MTF50 is the standard in optical and photographic measurements. The use of MTF9/10 (Rayleigh) is primarily used to determine the limit of human vision. The use of MTF0 (Dawes) is used to determine contrast at minimum separation...basically, detail is separated so minimally that, in a photographic context, an image of parallel white and black lines would look like pure gray. An image measured at MTF9 would still look almost purely flat gray, and the most dominant source of pixel deviation is going to be noise, rather than detail.

The continued STANDARD use of MTF50 is to ensure that fine detail transmitted by a lens at lower contrast is still resolved.

There are certain specific use cases where measuring at MTF0 are useful. Under extremely high powered magnification, the bright points of starlight can against the black backdrop of night sky can be fully resolved. A very, very high magnification (and truly excellent seeing or the use lucky imaging over an extended period of time) is required in order for a sensor to actually render the complex Dawes diffraction spots of stars that are not actually fully separated. You also usually need to significantly oversample the stars to get any kind of meaningful result as well.

In a standard photographic context, MTF0 is pretty much meaningless. MTF9 means contrast is so low that noise is going to be the dominant deviation, and therefor is still primarily useless for measuring photographic equipment. Most optical test benches, most reviewers, most mathematical formulas that you can find online, are based on 50% contrast.

 

[Lee Jay]

Actually, MTF50 is the standard in optical and photographic measurements. The use of MTF9/10 (Rayleigh) is primarily used to determine the limit of human vision.

Or, say, telescope resolving power. This discussion is about resolving power, and MTF50 is way too low for that.

The use of MTF0 (Dawes) is used to determine contrast at minimum separation...basically, detail is separated so minimally that, in a photographic context, an image of parallel white and black lines would look like pure gray. An image measured at MTF9 would still look almost purely flat gray, and the most dominant source of pixel deviation is going to be noise, rather than detail.

The continued STANDARD use of MTF50 is to ensure that fine detail transmitted by a lens at lower contrast is still resolved.

I disagree. MTF50 is used to represent the lower limit of "pixel sharp". The catch is, you don't want "pixel sharp". If your images are sharp at the pixel level, then you are undersampling your optics, and therefore throwing away detail.

In a standard photographic context, MTF0 is pretty much meaningless.

I agree.

MTF9 means contrast is so low that noise is going to be the dominant deviation, and therefor is still primarily useless for measuring photographic equipment.

I disagree. In fact, I've shot and extracted detail below MTF5. Of course, how far you can go here depends greatly on the noise in the image. That's why stacking allows you to pull out much more detail than you can without stacking - the effective exposure is dramatically longer and therefore the noise is dramatically smaller.

At lower ISOs, MTF9 is perfectly reasonable as a rough limit on resolving power. You only really need MTF50 when you are at extremely noisy exposures or, when you want "pixel sharp" as your target which, as I said above, you shouldn't.

 

[jrista]

MTF9 means contrast is so low that noise is going to be the dominant deviation, and therefor is still primarily useless for measuring photographic equipment.

I disagree. In fact, I've shot and extracted detail below MTF5. Of course, how far you can go here depends greatly on the noise in the image. That's why stacking allows you to pull out much more detail than you can without stacking - the effective exposure is dramatically longer and therefore the noise is dramatically smaller.

At lower ISOs, MTF9 is perfectly reasonable as a rough limit on resolving power. You only really need MTF50 when you are at extremely noisy exposures or, when you want "pixel sharp" as your target which, as I said above, you shouldn't.

I'm guessing your MTF5 images were in an astrophotography context? When it comes to astrophotography, I agree. But, I also think that is a pretty radically different field than regular photography. With AP, we have a whole lot more options and means of reducing noise, which makes resolving ultra low contrast detail (say, the surface of the moon, or cloud band detail in Jupiter) vastly more viable.

When it comes to "terrestrial" photography, I think there are still too many things working against the ability of camera systems to meaningfully resolve detail that MTF9 is more likely to result in the question: "Is that noise or detail?" than it actually results in a meaningful result. There may be a few occasions where you can resort to stacking for terrestrial photography as well...still scenes for example, maybe landscapes in certain circumstances. But for the most part, I'd say such low contrast scenarios are more viable for very niche types of photography...namely astrophotography.

I myself have probably resolved detail around MTF9. I use lucky imaging to image the moon, and I've tried Jupiter a few times. I simply don't have the focal length required...for my sensor pixel sizes, I need somewhere between 8,000mm and 10,000mm...and ultimately, resolving power in a telescope is related to the diameter of the aperture...so were talking about a 12-14" telescope with a 3x to 5x barlow. THEN, yeah, then were really talking about proper oversampling and some major resolving power of very low contrast detail. (Plus, the way we process such data, were not talking about a single frame...it's usually hundreds or thousands of frames integrated together using Sigma-clipping and superresolution algorithms.)

 

[Andyx01]

Here is a way of calculating the effective extra reach or resolving power of a crop body versus FF, which will amuse the geeks among us...

MTF is not a measure of resolution, it is a measure of a lenses ability to transmit contrast of the original scene through a lens. That's all it measures.

If what you really want is true image resolution, then there are two correct ways of measuring it. The first is a measurement of the performance of the lens itself. This measurement can be found in using something like a USAF resolution test chart and a microscope to perform aerial inspections of the image of the chart after it has passed through the lens.

Problems with this approach include difficultly in setting up a test bench (it's not easy at all). Additionally, the final result will not include measurements of chromatic aberrations, image distortions, field curvature, etc. The most important missing element in this kind of test is the exclusion of the imaging system itself. However, if the question is about real optical resolution, this test will give you the right answer.

Which leads to the second way of measuring, well, actually calculating real image resolution. This is diagnostic and very simple to perform. Simply take the number of image points ("pixels") in your file, divide by the size of your sensor (in millimeters), and divide by two. This number will represent the number of Line Pair per Millimeter (the measure of the ability to go from one white line to one black line) that your sensor can resolve.

This is rather interesting in that you can quickly see that a Canon 7D 18 mpixel sensor is capable of resolving 116 line pair per mm. The Canon 5D MkII is capable of 78 lppmm. As a comparison, Sony's 36mpixel FF sensor is capable of resolving 102 lppmm and Phase One's monster 80mpixel IQ180 returns 97lppmm.

Taking this a step further, look carefully at the physical limits (as in optical physics) of optical resolution, as measured in lppmm. You see that at f/2, an optically correct lens will return 695lppmm in the center of the scene where the light's wavelength is 589.3mu (green). At f/11, an optically correct lens will return 123lppmm, dropping off to 92lppmm at 25 degrees off-axis tangential.

Looking at this over the years, I have come to realize there is seldom a lens in-capable of resolving so poorly that a sensor (or old film for that matter) could out-resolve the lens. Sure, there are other optical effects, but we are talking pure resolution here. Nothing more.

Think about this for a moment. MTF does _not_ measure optical resolution. While useful, it does _not_ tell the story of resolution, no matter how much "math" you throw at it. Secondly, and perhaps most interestingly, optical physics show diffraction limited resolution at f/11 EXCEEDS currently manufactured sensors ability to return that resolution in all cases.

Rather shocking, don't you think?

Back to the original poster's point: Canon's 7D sensor outresolves (using the correct application of the word "resolution") the FF sensors from any manufacturer. It does so, however, for reasons other than those that were brought up.

Correct application of rational thought and real world science can help us properly understand and identify the errors and misleading comments widely published by marketeers and critics of optical imaging systems.

You see, if Canon had made a FF camera with 7D's pixel density, then there wouldn't be any questions - FF wins, period.
Nikon did that with their D7000/D7100 vs D800/D810. You crop FF and you get almost exactly what the pre-cropped D7000/D7100 produce. In fact, you can just shoot in DX mode on D800...
So there is no global question about crop winning anything vs FF other than price.
The question is why Canon didn't make that 46mp FF camera?
Was it because consumers didn't ask for it? - Maybe.
Why didn't they ask for it? - Because they are too ignorant in How-It-Works department. They are affected by this ... More_Pixels=More_noise=Bigger_Files=My_Old_iBox_Can't_Handle_it=Crop_Is_Just_As_Good_It's_Just_Magically_Different_Because_Size_Doesn't_Matter=I_Better_Buy_An_Overkill_Lens_For_My_Crop_And_Not_Use_The_Rest_60%_Of_It=I_Don't_Need_That_Much_Pixels_But_Digital_Zoom_Is_Bad#!@ERROR*2&$ERROR...System_Reset~Hello_Micro_Four_Thirds=Must_Buy_35-100/2.8_Because_It_Is_70-200/2.8 ... virus .

Lmao, +1

You forgot the part about how the equipment doesn't matter, and a good photographer could take a photo better than the hubble deep field with their iPhone.

 

[Lee Jay]

I'm not saying MTF5 is ordinarily achievable with regular photography. I'm saying that, or maybe MTF9 should be used as "extinction". MTF50 is so far below extinction that's it's not even close. If that were the case, anything below MTF 0.5 on the lens charts would be so blurry that it might as well be unrecognizable at the pixel level, and that simply isn't the case. Look at the 16-35II, for example. Something like half of the MTF chart is below 0.5 (MTF 50) at 30lp/mm!

http://www.usa.canon.com/cusa/consumer/products/cameras/ef_lens_lineup/ef_16_35mm_f_2_8l_ii_usm

 

[Lee Jay]

Here's what Imatest says: http://www.imatest.com/docs/sharpness/

"MTF50 or MTF50P are good parameters for comparing the sharpness of different cameras and lenses for two reasons: (1) Image contrast is half its low frequency or peak values,hence detail is still quite visible. The eye is relatively insensitive to detail at spatial frequencies where MTF is low: 10% or less."

I agree, fully, with that. MTF50 is about "sharp" ("detail is still quite visible") while MTF10 or so is about "extinction" (the limit of resolving power). I use MTF5 for stacked astro videos, and 10 for regular low-ISO (say, 400 or less, give or take, on SLRs) photography.

 

[jrista]

Yeah, I agree with that, MTF50 for sharpness, MTF10 for extinction. When it comes to testing lenses, though, and I think still viable for performing simple relative comparisons like Alan is doing, MTF50 is fine. Not only that, it is VERY easy to get lens test results in MTF50...it is significantly more difficult to find them performed at lower contrast levels.

Alan's little calculation is just a basic means of relatively comparing cameras with lenses on them, that's all. And I think it works well for that.

 

[Lee Jay]

Yeah, I agree with that, MTF50 for sharpness, MTF10 for extinction. When it comes to testing lenses, though, and I think still viable for performing simple relative comparisons like Alan is doing, MTF50 is fine. Not only that, it is VERY easy to get lens test results in MTF50...it is significantly more difficult to find them performed at lower contrast levels.

Alan's little calculation is just a basic means of relatively comparing cameras with lenses on them, that's all. And I think it works well for that.

It's almost impossible to get lens test results of any sort, including MTF50. We only have system tests, and those include the AA filter.

As I showed, even Alan's calculation is self-inconsistent, at least when comparing DPReview data to DPReview's reporting of DxO data.

 

[jrista]

Yeah, I agree with that, MTF50 for sharpness, MTF10 for extinction. When it comes to testing lenses, though, and I think still viable for performing simple relative comparisons like Alan is doing, MTF50 is fine. Not only that, it is VERY easy to get lens test results in MTF50...it is significantly more difficult to find them performed at lower contrast levels.

Alan's little calculation is just a basic means of relatively comparing cameras with lenses on them, that's all. And I think it works well for that.

It's almost impossible to get lens test results of any sort, including MTF50. We only have system tests, and those include the AA filter.

As I showed, even Alan's calculation is self-inconsistent, at least when comparing DPReview data to DPReview's reporting of DxO data.

I think your overcomplicating things. Pick a source, any single source, for lens data. DPR is fine. Then, compare any two systems using only that data. That's all that's necessary. Alan's formula is only for simple comparisons. It isn't intended to be highly accurate, just accurate enough to demonstrate what you should generally be expecting, performance wise, in real life. So long as your data source for lens (or lens+sensor, as they really are) tests is consistent, whether DPR's data matches DXO data or anyone elses data doesn't matter.

 

[Lee Jay]

Yeah, I agree with that, MTF50 for sharpness, MTF10 for extinction. When it comes to testing lenses, though, and I think still viable for performing simple relative comparisons like Alan is doing, MTF50 is fine. Not only that, it is VERY easy to get lens test results in MTF50...it is significantly more difficult to find them performed at lower contrast levels.

Alan's little calculation is just a basic means of relatively comparing cameras with lenses on them, that's all. And I think it works well for that.

It's almost impossible to get lens test results of any sort, including MTF50. We only have system tests, and those include the AA filter.

As I showed, even Alan's calculation is self-inconsistent, at least when comparing DPReview data to DPReview's reporting of DxO data.

I think your overcomplicating things. Pick a source, any single source, for lens data. DPR is fine. Then, compare any two systems using only that data. That's all that's necessary. Alan's formula is only for simple comparisons. It isn't intended to be highly accurate, just accurate enough to demonstrate what you should generally be expecting, performance wise, in real life. So long as your data source for lens (or lens+sensor, as they really are) tests is consistent, whether DPR's data matches DXO data or anyone elses data doesn't matter.

Even though one got 1.2x and one got 1.5x with basically the same pixel sizes? That's a pretty enormous range when 1.0-1.5 is what one would get without any data at all and just assuming either a horrible lens (1.0) or a "perfect" lens (1.5).

 

[jrista]

Yeah, I agree with that, MTF50 for sharpness, MTF10 for extinction. When it comes to testing lenses, though, and I think still viable for performing simple relative comparisons like Alan is doing, MTF50 is fine. Not only that, it is VERY easy to get lens test results in MTF50...it is significantly more difficult to find them performed at lower contrast levels.

Alan's little calculation is just a basic means of relatively comparing cameras with lenses on them, that's all. And I think it works well for that.

It's almost impossible to get lens test results of any sort, including MTF50. We only have system tests, and those include the AA filter.

As I showed, even Alan's calculation is self-inconsistent, at least when comparing DPReview data to DPReview's reporting of DxO data.

I think your overcomplicating things. Pick a source, any single source, for lens data. DPR is fine. Then, compare any two systems using only that data. That's all that's necessary. Alan's formula is only for simple comparisons. It isn't intended to be highly accurate, just accurate enough to demonstrate what you should generally be expecting, performance wise, in real life. So long as your data source for lens (or lens+sensor, as they really are) tests is consistent, whether DPR's data matches DXO data or anyone elses data doesn't matter.

Even though one got 1.2x and one got 1.5x with basically the same pixel sizes? That's a pretty enormous range when 1.0-1.5 is what one would get without any data at all and just assuming either a horrible lens (1.0) or a "perfect" lens (1.5).

Pixel size alone is not the sole factor in determining sharpness and microcontrast. AA filter, as you said yourself, plays a role, among other things. You can resort to manufacturer mathematically generated MTF charts if you want a more consistent source of data. Those charts tell you a wealth of information...but you don't have to use it all. You can choose to use the f/8 or max aperture data, and you can pick a range from center to corner for your data points.

 

[Lee Jay]

Pixel size alone is not the sole factor in determining sharpness and microcontrast. AA filter, as you said yourself, plays a role, among other things. You can resort to manufacturer mathematically generated MTF charts if you want a more consistent source of data. Those charts tell you a wealth of information...but you don't have to use it all. You can choose to use the f/8 or max aperture data, and you can pick a range from center to corner for your data points.

Yeah, but the 5DIII is nowhere close to this much better than the 5DII as far as resolution goes.

Canon's calculated MTF charts are a major problem since they don't include diffraction. You'd have to adjust all of Canon's data to include that, somehow, and it's not even clear how exactly they exclude it.