Size matters, but just how?

It is generally accepted that, to speak rather imprecisely, a camera with a larger sensor offers the potential of "better image quality" than one with a smaller sensor.

This should come as no surprise. Information theory teaches us that information and energy are intimately related. For a given scene (with a certain distribution of luminance), assuming an aperture of f/2.8 and a shutter speed of 1/200 sec, a 43 mm sensor ("full frame 35 mm") receives about 31 times the energy in an exposure than the 7.7 mm sensor in my new Panasonic Lumix DMC FZ200.

But, in a more specific way, just how does this happen?

Well, the larger real estate of the larger sensor can be deployed in two ways (between which the designers must trade off):

• More sensels, and thus greater potential for image resolution.

• Greater sensel area, and thus the potential for better signal-to-noise ratio at any sensor luminance.

Now, in processing the sensor output, we can apply algorithms that improve the "noise" behavior of the image, generally at the expense of deterioration of some other image quality property. Some of this happens in the camera, and some of it may happen in post-processing.

Now, let's look at actual happenings. We might expect the image from a camera with a 43 mm sensor (perhaps a Canon EOS 6D) to be "better in overall quality" than an image from a camera with a 27 mm sensor (perhaps a Canon EOS 70D).

But in just what way? How does this difference manifest itself, for example, on a print at a size of 12" x 8"? What is the difference that the viewer observes? And is the impression of a "better image" in part because of properties that the viewer may not be able to recognize but which nevertheless affect the overall "niceness" of the image?

Best regards,

Doug

I am glad you are starting this subject which I consider starting myself. I can't wait to read what your conclusions are.

Having a non-technical approach to photography beyond the understanding of the functions of knobs and dials, I've never really considered sensor size as a decision maker for taking a shot. I have cameras that I use equally as often with a range of sensor sizes and in my ignorance, I used to choose a camera according to the size of the final print. Then I discovered that all my produce doesn't really warrant such a consideration. The artistic considerations seem to go beyond the necessity to have more information. And by the time I've finished and displayed, any necessity for absolute clarity is lost.
Certainly this is only my reflection and will not necessarily bare witness to others on this site. My house is full of prints. No-one to my knowledge has ever commented on the size of the sensor or related matters. So, as the sample of viewers goes, and they are, admittedly, not all that discerning, it doesn't seem to make a lot of difference to them. Of course, they are somewhat inhibited in their comments since they know if they comment on the sharpness they have to suffer the pain of a swift kick to the testicles or a nipple pinch, whichever seems appropriate.
As for the term 'niceness', again, the word is banned in my house. Isn't there a better word you can come up with, Doug. What about a negative scale like 'shitness'. At least it's got some punch. Niceness is so .... Erh ... Nice!

Hi, Tom,

Aw, you are always so—well, not really.

Best regards,

Doug

Doug Kerr said:

Hi, Tom,

Aw, you are always so—well, not really.

Best regards,

Doug

Click to expand...

Don't go away, Doug.

So ...... what? Don't leave me hanging.

I thought I added to the subject seriously. In a nutshell, I was saying, to my viewers, size doesn't seem to matter. Especially to my wife, who see my pictures all day long and says how nice they are.

Cheers
Tom

Print size obviously matters or classical painters would not have gone through the trouble and expense of painting on huge canvases. It was actually one of my biggest surprises when I visited the Louvre museum for the first time at a child: to realize how big paintings were which I only knew from my classbooks. For example, this is about 5x7 meters:


In context:


This intimate portrait is 2x1.5 meters:


in context:


The size of the prints matters just as much for photography. Think about people like Hiroshi Sugimoto or Andreas Gursky for example. Why should we pretend otherwise? People insisting that size does not matter usually do not want to go through the trouble and expense that producing large prints necessitates. Yet, this is the easiest way to convince your viewers to take you seriously, especially in a world already saturated with screens and pictures. Why should we restrain from printing big?

Jerome Marot said:

The size of the prints matters just as much for photography. Think about people like Hiroshi Sugimoto or Andreas Gursky for example. Why should we pretend otherwise? People insisting that size does not matter usually do not want to go through the trouble and expense that producing large prints necessitates. Yet, this is the easiest way to convince your viewers to take you seriously, especially in a world already saturated with screens and pictures. Why should we restrain from printing big?

Click to expand...

A very good demonstration Jerome, I can't fully agree more!

I should also make clear that big prints may have other uses than just impress by sheer size. For example, in the works by Gillian Wearing about masks, the picture can only be understood by look at the details. This is the most known picture in the series: "Self Portrait at 17 Years Old". Gillian Wearing was 40 at the time. The print is about life-size.

As a small print:


At that size, it simply looks like a 17 years old girl. It is designed to fool us, even the curtain behind the subject reminds us of identity pictures.










But when printed at the intended size, we can more closer and discover from the details that she is actually wearing a mask:


Here the size of the prints is necessary to communicate the message: we are all wearing masks, presenting an identity which is not our true self. Only by coming close enough to the picture/person can the viewer perceive the small details that reveal the real person behind the mask.

Hi, Tom,

Tom dinning said:

Don't go away, Doug.

So ...... what? Don't leave me hanging.

Click to expand...

I didn't propose to give the answer—the original post was to raise a question in hopes that others here could shed some light on the issue.

I thought I added to the subject seriously. In a nutshell, I was saying, to my viewers, size doesn't seem to matter. Especially to my wife, who see my pictures all day long and says how nice they are.

Click to expand...

Well, I hope you don't chastise her for saying "nice".

But you response was otherwise well taken.

Thanks.

Best regards,

Doug

Hi, Jerome,

Your illustrations are very valuable.

Note however that I was speaking of sensor size (although of course the matter of print size is certainly part of the overall matter).

Best regards,

Doug

Hi, Jerome,

Jerome Marot said:

I should also make clear that big prints may have other uses than just impress by sheer size. For example, in the works by Gillian Wearing about masks, the picture can only be understood by look at the details. This is the most known picture in the series: "Self Portrait at 17 Years Old". Gillian Wearing was 40 at the time. The print is about life-size.

Click to expand...

Now, returning to my original quest, what specific describable technical properties would the use of a larger sensor confer on such an image that would increase its "goodness" in the context of such a larger presentation?

Best regards,

Doug

Doug Kerr said:

Now, returning to my original quest, what specific describable technical properties would the use of a larger sensor confer on such an image that would increase its "goodness" in the context of such a larger presentation?

Click to expand...

The 3 photographers I cited (Hiroshi Sugimoto, Andreas Gursky and Gillian Wearing) all use silver based photographic film (in the case of Gillian Wearing, this is true for the pictures cited, not necessarily for more recent pictures). Your question has a relatively trivial answer for film, since its resolution does not depend on exposure size, larger film size simply convey more information.

This answer is trivial, but is only an approximation. Keeping film flat on large surfaces is difficult, so larger formats lose some resolution compared to smaller format (at least for roll film). Designing lenses is more difficult for larger image circles, so small format lenses can have a higher resolving power. Large format lenses are usually used at apertures where diffraction comes into play as a real limitation. And, of course, slower or more modern film can have a higher resolving power.

Before we discuss digital format, I would like to give the approximative dimensions of common sensor (or film) sizes:


Name.......... Dig. size..........Film size (mm)
1/2.5"..........5.76x4.29..................
1/1.8"..........7.18x5.32..................
1/1.7"..........7.6x5.7....................
Disc Film..............................10x8
Minox..................................11x8
Nikon 1.........13.2x8.8...................
110....................................17x13
4 thirds........7.3x13....................
APS.............23.6x15.7..............25.1x16.7
18x24...........24x18....................
"full frame"....36x24..................36x24
MFD 1...........44x33...........................
LeicaS ........ 45x30...........................
MFD2............49x37...........................
MFD3............54x40.4.........................
645....................................56x41.5
6x7....................................69x56
4x5"...................................126x100
8x10"..................................240x180



(MFD1, 2 and 3 are the sensor sizes of Digital medium format backs).



(and for bigger film- different scale)



One should notice that digital sensors are generally smaller and that, if we exclude cell phone tiny sensors, the rate between the biggest and smallest sensor is smaller for digital than for film, especially if we consider surface and not linear dimensions.

Just to animate the discussion a little.

The choice of the sensor size can be application-specific.
I had a few (raw) thoughts on this.

Best regards,
Michael

Hi, Jerome,

Thanks for the nice reference.

I myself have always liked Kleinbild as the name for that frame size bigger than which Canon shooters can (so far) hardly grasp.

Best regards,

Doug

Doug,

I believe one has to add size of "print" or screen to the discussion as Jerome so elegantly pointed out with his excellent examples. At 8x10 size or about, the only serious differences, as Tom Dinning discovers, is the nature of the picture.

We cannot stand a crackle or pop in a movie sound track, but an underexposed well composed sequence is considered "Artistic" and we always give it a pass. I think the same is true of the still picture. Most of the concerns we have for cameras that give us small prints are not relevant to enjoying the picture or getting awarded a prize!

Unless the need is for detail rich images, most modern cameras do fine, except some require walking to get the picture in the frame.

So I'd think this discussion is best framed as the relevance of sensor size to getting pictures of certain classes of size and detail and viewing distance.

A 6MP camera can be used for billboard pictures to be seen across the freeway! But you know that!

Asher

Asher and Michael both make interesting points: if we want to assess the differences between one camera and another, we should assess it in relation to our purpose of making pictures. I wanted to raise this point later and I shall come back to it. In any case, I am not interested in studying the camera per se. I am interested in studying a photographic process: camera, subject and lens to realize prints. We will see that limiting ourselves to real photographic cases is important when we want to assess some effects like depth of field or dynamic range.

To study the case of digital sensors, I suggest to compare two cameras with similar resolution but different sensor sizes, the Nikon D800 and the Leica S. As an exercise, I suggest to study what happens if we would make the D800 bigger to match the sensor size.

The two cameras have similar resolution (D800: 36.3 mpix, S: 37.5 mpix), but one sensor is 36x24mm and the other 45x30mm. The linear dimensions are 1.25 bigger (25%) for the S. I will imagine that we have an expanding machine that can blow the D800 25% bigger in every dimension, creating a D800+. What happens?

The D800 dimensions are 146x123x81.5mm, the D800+ is 182.5x153.5x101.5mm.
We still have 36.3 million pixels, they just get bigger (we could have chosen to get more pixels of the same size). Dynamic range for highlights being roughly proportional to the surface of pixels (for a given technology), we would increase dynamic range. This is irrelevant for photographic practice, since the D800 dynamic range is already sufficient for photographic subjects (we shall come to that later).
The weight is multiplied by the cube of 1.25. This is often overlooked when scaling objects, the weight is proportional to volume, i.e. dimensions to the cube. Our 1000g D800 becomes a 1953g D800+.


Since we are interested to a complete system, we will fit that D800 with a standard lens, a 50mm f/1.8G, 7 lenses in 6 groups, and see what happens. The lens becomes a 62.5mm. The dimensions increase from 72x52.5mm to 90x65mm. Weight increases (cube again) from 185g to 361g. Aperture stays identical at f/1.8, because it is a dimensionless number (the ratio of 2 dimensions).

Let us compare the D800+ with its 62.5mm lens to the Leica S.

The D800+ is bigger than the Leica S (182.5x153.5x101.5mm versus 160x120x80mm) and heavier (1953g versus 1410g). We should expect that, since Leica does not use a blow up machine, but uses standard components for the electronics (processor, memory, etc...). The only things which need to be bigger in the S are the sensor and the mechanics (shutter, mirror box, mount).

The real difference are the lenses. Leica standard lens focal is a bit longer at 70mm versus 62.5mm. But Leica standard lens is also:
-slower: f/2.5 versus f/1.8
-uses one element more: 8 versus 7
-is longer (93mm versus 65mm)
-has the same diameter, even if it is slower (90mm for both)
-is much heavier at 740g versus 361g.

Leica lenses for the S series are particularly complex and heavy, but we would find a similar situation for Hasselblad or Phase one: medium format lenses are slower and more complex than their small format counterparts, even taking account of linear scaling. The reason they are so is interesting and shall be the subject of the next post.

Hi, Asher,

Asher Kelman said:

I believe one has to add size of "print" or screen to the discussion as Jerome so elegantly pointed out with his excellent examples. At 8x10 size or about, the only serious differences, as Tom Dinning discovers, is the nature of the picture.

We cannot stand a crackle or pop in a movie sound track, but an underexposed well composed sequence is considered "Artistic" and we always give it a pass. I think the same is true of the still picture. Most of the concerns we have for cameras that give us small prints are not relevant to enjoying the picture or getting awarded a prize!

Unless the need is for detail rich images, most modern cameras do fine, except some require walking to get the picture in the frame.

So I'd think this discussion is best framed as the relevance of sensor size to getting pictures of certain classes of size and detail and viewing distance.

Click to expand...

That is an important matter.

But that was not my inquiry, which was:


Not "how did it get that way", or "why did we choose to make it that way", or "why did we want a print that size anyway". Just, here are two prints. We feel that one looks better than the other. There must be one or several properties that are different between them as we have them in hand.
What might those properties be?

Best regards,

Doug

Doug Kerr said:

Just, here are two prints. We feel that one looks better than the other. There must be one or several properties that are different between them as we have them in hand.
What might those properties be?

Click to expand...

We understand your question, but the answer is very different if these two prints are relatively small (say, computer screen or A4 size) or relatively large (say, as large as a human adult). So which size are your prints?

And, please, do not come with the argument that large prints are the same as small prints, the viewing distance should just increase. The public does not do that. The effect of large prints in photography comes from the fact that they can be examined quite close and still look sharp.

Jerome Marot said:

the viewing distance should just increase. The public does not do that. The effect of large prints in photography comes from the fact that they can be examined quite close and still look sharp.

Click to expand...

Too bad but just true…
Billboards does not count as they aren't done for being tak sharp, but to be seen from large distances. They are usually printed at 15 dpi, imagine the dots size!

Doug Kerr said:

But that was not my inquiry, which was:

"We all 'know' that if we have two prints of the same size [let's say, 'pretty large'], of identical content, one from an image taken with a camera with a certain sensor size, and one taken with a camera with a larger sensor, the second image will typically 'look better'. What properties does the larger image have that makes it 'look better' ".​

Click to expand...

Hi Doug,

Let's assume for a moment that the premise that the image from a larger sensor camera will typically look better is true (which is not always the case).

It's hard to say without an actual example, but there are several possibilities for such differences. Resolution might be one, but only if the print size was large enough to show the limits of too few pixels for the output size.

Noise may be a factor, which is usually related to sensor design and 'well depth' that translates into different Signal to Noise (SNR) and dynamic range characteristics.

It's more likely that the system MTF is different, as a result of a combination of different lens (longer focal length and smaller aperture to equalize DOF), different sensel pitch and thus sampling density, and Bayer CFS color filters. Sometimes there is a difference due to the lack of an optical low-pass filter (OLPF).

Differences may also exist due to sub-optimal use of the equipment, and due to the post-processing of the images (both Raw conversions and tonemapping).

Cheers,
Bart

Nicolas Claris said:

Too bad but just true…

Click to expand...

There is nothing bad about it. This is one of the features which characterizes photography: one can fill a wall-sized print with a wealth of details. The effect for the viewer is overwhelming. Why should we not want that the public examines our large prints with a magnifier if that creates overwhelming emotions in their hearth? Whatever works, works.

Please note that this has also been tried by painters. Here is a link to a flash animation allowing one to zoom into a painting by Jean Brueghel: La Terre (1.5 billion pixels...).

Bart_van_der_Wolf said:

Noise may be a factor, which is usually related to sensor design and 'well depth' that translates into different Signal to Noise (SNR) and dynamic range characteristics.

Click to expand...

That is the simplified version, which ignores several characteristics of real semi-conductor sensels: sensels are supposed to be simple wells which convert photons to electrons and are only limited by thermal noise (also on the converters) and by the finite storage space they have for electrons before saturation. In this simple model, a sensor with tiny sensels can be just as good as a sensor with bigger sensels, provided that:
- we have enough signal to be reasonably above noise
- we make the "storing well" deep enough to avoid saturation.
In other words, for this simple model, the only problems that we can predict for tiny sensors is that they will blow up highlights more easily and will be noisier for low light levels.

The reality of semi-conductor sensels adds several other limitations to this simple model, which I am planning to discuss latter in the day with the help of a paper from Aptina.

Hi, Jerome,

Jerome Marot said:

We understand your question, but the answer is very different if these two prints are relatively small (say, computer screen or A4 size) or relatively large (say, as large as a human adult). So which size are your prints?

Click to expand...

18 in. × 12 in.

Best regards,

Doug

Hi, Bart,

Thank you for addressing my question. I think we are moving toward an answer.

Best regards,

Doug

Doug Kerr said:

18 in. × 12 in.

Click to expand...

So roughly A3 size: 30x45cm. I have some prints of that size on my walls, made with a 6 mpix camera at the time. They look just fine.

Edit and correction: I had a closer look. The pictures were taken with an EOS 300D and a small point and shoot. They look oversharpened. The one from the P&S are less sharp and visibly noisier in some cases. They are definitely not as good when examined close on as the A1 prints made with a latter A900 full frame camera. Still, the quality is good enough for that size when examined casually.

Doug Kerr said:

Thank you for addressing my question. I think we are moving toward an answer.

Click to expand...

And what would be your answer to that question? Surely, you have some ideas on the subject.

Hi, Jerome,

Jerome Marot said:

So roughly A3 size: 30x45cm. I have some prints of that size on my walls, made with a 6 mpix camera at the time. They look just fine.

Edit and correction: I had a closer look. The pictures were taken with an EOS 300D and a small point and shoot. They look oversharpened. The one from the P&S are less sharp and visibly noisier in some cases.

Click to expand...

Well, there you go!

It is curious that some that appear to be "oversharpened" still appear "less sharp"! Perhaps we are unsure what we mean by "sharp".

They are definitely not as good when examined close on as the A1 prints made with a latter A900 full frame camera. Still, the quality is good enough for that size when examined casually.
Sure.

And what would be your answer to that question? Surely, you have some ideas on the subject.

Click to expand...

No, I am completely mystified. That's why I posed the question of my colleagues.

Best regards,

Doug

Doug Kerr said:

Well, there you go!

It is curious that some that appear to be "oversharpened" still appear "less sharp"! Perhaps we are unsure what we mean by "sharp".

Click to expand...

The oversharpening is not general for all prints made with small sensor cameras. It was, on one hand, incompetence from my part and, on the other hand, due to the relatively low pixel count for the print size.

And yes: oversharpened pictures can appear less sharp. They have more "bite", but they still miss fine detail. It looks a bit like an analog TV picture with too much high-frequency boost, if you get my analogy...

Hi, Jerome,

Jerome Marot said:

The oversharpening is not general for all prints made with small sensor cameras. It was, on one hand, incompetence from my part and, on the other hand, due to the relatively low pixel count for the print size.

And yes: oversharpened pictures can appear less sharp. They have more "bite", but they still miss fine detail. It looks a bit like an analog TV picture with too much high-frequency boost, if you get my analogy...

Click to expand...

I understand.

Best regards,

Doug

On CCD and CMOS detectors.

As I said earlier, electronic detectors are limited by various sources of noise. A good summary of the subject is to be found here but you probably already know the essentials: basically, sensels are supposed to be simple wells which convert photons to electrons and are only limited by thermal noise (also on the converters) and by the finite storage space they have for electrons before saturation.

The realities of electronic detectors are a bit different. But before we go into details, let us look just how small the sensels really are. These are typical figures:

Canon EOS 300D (digital Rebel): 7.4 µm
Typical 50 or 60 mpix medium format: 6µm
Typical 24 mpix full frame: 6µm
Phase One IQ 180: 5.2 µm
D800: 4.7µm
24 mpix APS-C camera: about 3.9 µm
Canon powershot G9: 1.9µm
iPhone 5S: 1.5 µm
Nokia Pureview: 1.4 µm

But when we shrink a camera system and sensels, there is one think that we cannot shrink: the light. We don't take UV pictures with small sensors, so how big is the wavelength of visible light by comparison?

Red light: 0.75 µm
Yellow light: 0.58 µm
Violet Light: 0.4 µm

From the comparison between these figures, it should be clear that we are not too far from the limit. A sensel smaller than the wavelength of light would be the subject of different effects. Basically, it would act as an antenna.

Now, I would like to point you to the following document: http://www.aptina.com/news/FSI-BSI-WhitePaper.pdf
Basically, this document is a discussion of front side and back side electronic sensors by Aptina (a sensor manufacturer). I will use the figures for my explanations.

Turn to figure 1: this is the typical design of a front side sensor. From top to bottom, you have microlenses, a color filter array, a "light guide" in the oxide layers (where the wires with reading and command circuits go) and, deep inside, the actual detector to convert photons to electrons. Column 1 has no light guide, column 2 is standard and column 3 increases the capacity of the photodiode. This explains the progress in sensor design between generations: 2 loses less light than 1 and 3 increases the maximum storage capacity (so it won't burn highlights…).
Turn to figure 3 for schematics of a back side sensor (only used for very small sensels).

What are the problems with these real sensels? How do they increase when we decrease the sensel size?

The first problem is crosstalk: there is some signal leaking from one sensel to the other (directly as light or electronically) and the smaller the sensel, the bigger the leak. Since adjacent sensels treat different colors because of the bayer array, this crosstalk directly translate as less color separation.

The second problem is due to the complicated structure in front of the diode: a lens and a light guide. This is an optical system and, as such, has its own limitations. Typically, it can only see rays coming from a relatively narrow cone. This, in turns, restrict the maximum aperture of the lens of our camera (actually, this depends on the optical system of that lens, hence the search for so called telecentric lenses) and more when one is far away from the axis of the lens (i.e. the sensels at the side of the sensor will restrict more, one gets a kind of vignette effect). These effects are strongly increased with smaller sensors (they are quite noticeable with 4 µm sensels).

Back to photography practice: what difference will we see between a camera with large sensels (say a 6µm full frame camera) and one with smaller sensels (say a typical 4 µm APS-C or a typical 2µm point and shoot)? Easy:
-we will have less color purity (crosstalk)
-the smaller camera will exhibit more noise in dark zones (because the signal is smaller, but noise sources are relatively constant)
-the smaller camera will saturate earlier because the diode is smaller, hence has a smaller capacity (i.e. we will blow highlights)
-the smaller camera will restrict the design of the lens.

That last effect is too often ignored, but is quite noticeable on recent APS-C or µ4/3 cameras. Between 2 generations, lenses which used to work fine started vignetting like crazy. Of course, at the same time, the manufacturers introduced vignette correction to fool us… For point and shoot of cell phones, the manufacturers are silent in the limitation of the lens, but be insured that designing a more telecentric lens comes as an additional constrain which the optical engineer does not need to design a sharp lens. Basically, if you want a lens capable of being used with tiny sensels and, in addition, a large zoom range, a compact system and plastic mechanics for the system to be light and not too expensive, you should be surprised if it works at all. Part of the magic is bought with in-camera digital corrections, of course.

In our next and last series, we shall concentrate on the subject of lenses, limits due to diffraction and find out why large sensors leads to better bokeh.