Camera Sensors

Pixel size, Pixel counts, and Sensor size

Pixel Size

The size an individual pixel is has an important bearing on the final quality of an image. The larger a pixel is, the quicker it responds and gathers information. The smaller a pixel is, the more amplification is needed to generate the electrical signal required to provide the information the sensor supplies. As the amount of amplification increases, so does electrical interference known as 'noise', which degrades image quality. This is known as the SNR - signal to noise ratio. Noise manifests itself as different coloured pixels in areas of even colour or tone, and is most prevalent in low light level shots such as those taken indoors, or outside at night using colour. It is not so much of a problem in Black & White shots but does lead to 'speckling' of tonal values.

Generally each sensor has base sensitivity at which it works best. The larger the pixels the more sensitive a sensor will be. So the ISO ratings range a camera may have is often a reflection of the size of the pixels and how much amplification of the sensor signals is required. ISO ranges are provided by increasing the amplification - sensitivity - of the sensor signals. If a sensor has very good signal response at low amplification then it is possible to provide a wide ISO range with increased sensitivity and keep the SNR ratio low. 

So the bigger a sensor is, and the bigger the pixels are, the better the image quality will be. But a camera's size, and that of any lens it uses, is related to the size of it's sensor. The larger the sensor - the bigger the camera. As the cost of a sensor is also related to its size, this explains why the D-SLR's with their larger sensors, produce the best image quality - and also cost more to buy.  Please see  Digital Camera ISO settings  for further details

Dynamic Range

When you take a picture using a camera you are trying to record all the tones in a scene. It is the different tones, from light to dark, and their relationship to each other that together make up a recognizable image. There are 256 recordable tones, from pure white, to pure black in an Jpeg 8bit file image. Under dull or overcast conditions recording the full range of tones is not usually a problem. But in bright, high contrast conditions it is. In these situations and depending on how the scene has been metered by the camera -  see  Metering   - either shadow or highlight detail can be lost. Film generally has a wider dynamic range than most digital camera sensors at present, but even it cannot cope in all situations, whilst print film is better than slide. 

See Image Exposure  and  Dynamic Range Assessment

The reason the dynamic range of sensors is not as wide as film has everything to do with the way the different mediums work. Film is composed of tiny grains of chemicals which change colour in response to light. When each grain has absorbed all the light it can up to it's maximum it can't absorb any more, even if the exposure continues as other grains continue to absorb light in areas where there is less. This means that it is easier to capture a wider range of tones in high contrast scenes. 

With sensors it's different. A pixel continues to absorb light even past maximum white light, until the exposure ceases, it can't 'switch off' before. This leads to two problems. One is that in high contrast scenes there is light 'spillage', where the maximum white light readings tend to 'spill over' and affect other adjacent pixels so they produce a much 'whiter' colour reading than they should, and affect other pixels in turn. This in turn produces what is known as 'burnt out highlights', areas where all detail is lost because it's too bright. The answer is to reduce the exposure to prevent this. This in turn leads to the second problem, that in areas of low light the pixels don't absorb enough light and this leads to loss of detail and contrast.

Various software algorithms are now being developed to try and address the problem by reducing the light spillage in bright areas and at the same time boost the contrast in dark and shadow areas. This is being introduced in various DSLR cameras as Dynamic Range Optimization, Highlight Tone Priority, or something similar.

ISO Range

The larger the pixels are on a sensor, the faster it gathers information - the more sensitive it is - and the more able it is to record the greatest range of tones in a scene. A benefit of this is that it can do this over a greater range of ISO speeds than a sensor with smaller pixels, and maintain image quality. As sensor pixels get smaller in size, and thus require longer exposure to the light to gather the required information, so the ISO speeds get slower, and the ISO range they can deliver good images at gets narrower. 

A good measure of a digital camera's ability is thus often it's ISO range. A wide range indicates a capable sensor, a narrow one less so. A low base ISO also indicates a sensor with small pixels and slow light gathering powers. Usually the two are connected. A sensor with small pixels will have low ISO rating and a narrow range. Many DSLR's have large pixel sensors with ISO ranges from 200-3200, or 100-1600. Digicams however often only have narrow ranges such as 100-400, or 50-200. 

The lowest ISO figure that a digital camera has is usually the rating at which the best image quality exists that the sensor can produce. But it is not the case that the deterioration in image quality, as ISO speeds increase, is uniform. One camera might produced images across a wide ISO range where there is only limited degradation, whilst another might have huge and unacceptable differences across a narrow one. Often this is down to how well the software that has been written for the camera can interpret the information it receives from the sensor and what level of noise reduction is used. High levels of noise reduction can result in reduced image detail.

Pixel Counts.

At the present time, most sensor sizes are made with differing numbers of pixels, the smallest being 5mp. The 1.2/7", 1.1/8" and 2/3" types are used in digicams and the maximum is currently 8mp for the 2/3rds size and 7mp for the smaller ones. The 4/3rds, APS-C and 35mm ['full frame sensor'] types appear in the D-SLR's and counts range from 4.7mp through to 10/12/14mp, with 12mp/21/24mp for the full frame types. 

One of the main problems camera makers and chip designers face is that as more pixels are fitted onto a sensor they must as a result be smaller. As pixels get smaller the signal to noise ratio increases, and image quality decreases. So the challenge is to put as many pixels on a chip as possible whilst keeping image noise to a minimum. Another emerging problem is that high sensor counts are exceeding the resolution levels of the lenses used with them.

In general terms, the bigger a sensor is, the better the image produced will be. But for any one sensor size, while a higher pixel count will produce a finer detailed image, able to be printed at a larger size without pixels showing, in some cases the image quality may be lower due to increased noise due to the smaller size of the individual pixels. It is for this reason that the top of the range 'prosumer' digicams use 2/3" sensors, whilst DSLR's use much larger types. Generally most DSLR images have little or no noise at all, in comparison with even the best 'prosumer' digicam. As you can see from looking at the previous chart, even a 2/3" sensor is quite small and its pixels about 1/4 to 1/6 the size of those on the 'APS-C' sensors.'  We are now of the opinion that pixel counts for many of the chip sizes have not only reached their limit, but gone too far, and overall image quality and usability - resulting in higher noise levels and too low ISO's - has been sacrificed as a result.

Making comparisons

The ordinary digital camera user tends to get very confused, as do we from time to time, trying to work out and understand the relationship between the actual size of individual sensors, the number of pixels that are fitted on to them, the size that the actual pixels on the different sensors are, and what this means in actual user terms for the digital camera user, in respect to reproduction and printing resolutions.

For at the end of the day the biggest single factor would appear to be the actual size of pixels on a sensor. A larger number of pixels does enable improved image quality in the sense that an image can be produced at a larger size without looking pixelated i.e. displaying jagged edges. And this does mean in turn that at any given size image detail will look finer. But there is a limit as to how much detail the human eye can distinguish. Once you move beyond that point extra pixels are superfluous. More to the point they are unnecessary and in some ways counter productive since they increase file sizes for no particular benefit and take longer to process.

There are two main comparisons that can be made with regard to sensor and pixel size. There's resolution, which is concerned with the pixel count. And noise level, which is mainly a result of pixel size. A clear dividing line can be drawn between the smaller sensors used in digicams and the larger sensors used in DSLR's and this is as a result not of resolution but noise levels.  

Noise level is directly connected to the actual size of the pixels on a sensor. Various methods are employed to reduce excessive noise using software algorithms where possible, but these can have a degrading influence on overall image quality. The best solution is not to have high noise levels in the first place and this is achieved by using the largest pixel size possible. In order to illustrate the difference that exists we have compiled the following table and diagrams which illustrate the size of pixels on various sensors in relation to each other. 

We don't know the size of, and can't measure, the actual pixels on a sensor. What we can do is make comparisons by calculating the area allowed for each pixel and thus the pitch between them, i.e. from centre line to centre line. It's not totally accurate as there is space around each pixel for wiring etc. Nevertheless it's valid as a comparison, and easy to do by taking the width or depth of the sensor, and dividing it by the number of pixels located along it. 

So for the common APS-C sized 6mp sensor used by Pentax, Nikon and Konica-Minolta in many of their DSLR's it's simply a matter of dividing the sensor width, 23.5mm by the number of horizontal pixels, 3008. A figure of 7.8 microns results. A micron is a millionth of a metre. It is usually written as µm.

Here is a table listing a range sensor sizes and pixel counts with the resulting pixel size in microns. Some relate to specific well known cameras with unique sensor sizes or pixel counts, and where sensors are common, to a general sensor size/pixel count.

This table is interesting and clearly shows why so many DSLR makers stuck to using the 6mp APS-C sensors in their cameras for so long. Only the much more expensive 12mp full frame DSLR's from Nikon and Canon have larger pixels. Although larger pixel count cameras have been introduced the main problem has been one of keeping noise levels down to near that produced by the 6mp sensors. And a lot of R&D has been spent achieving that. If you consider that the 6mp APS-C is 2.25 times smaller than a full frame sensor it's easy to see how the 12mp full frame sensors have higher counts with bigger pixels.

It also shows why digicam makers choose 2/3" chips for the 5mp prosumer digicams, the pixels were almost the same size as the then current 3mp digicams using smaller chips. The subsequent increase in pixel count to 8mp was not a good idea, as we have said before. But seeing is believing as they say, so below are some diagrams illustrating a few sensor pixel sizes in relation to each other.

Eos 5D  12.8mp			6mp APS-C
Eos 1Ds11 16.7mp			Eos 350D/20D 8mp/8.2mp
10.2mp  APS-C			12.2mp APS-C
Olympus E300/E500 8mp			3mp 1.1/8"
5mp 2/3"			8mp 2/3"
6mp 1.1/8"			9mp 1.1/8"

Resolution and Moire

The other main comparison most users make between digital cameras, whatever their type, is sensor count. If increased sensor count has a detrimental effect on image noise because of smaller pixel size there is one area where an advantage is gained, increased resolution, and thus finer image detail.

Higher pixel counts also help to offset to some extent problems that exist with sensors, digital artefacts. There are two. One we have already mentioned, aliasing, occurs as a result of colour interpolation. The other, moire, is a result of the highly structured nature of sensor pixels clashing with patterns in a scene. Finely woven fabric is one, but any regular repeating pattern is liable to cause it. It happens when the frequency of the pattern is smaller than the frequency of the pixels. It is more likely to happen with distant scenes and objects where the regular repeating patterns are smaller in size in relation to the number of pixels on a sensor. 

To overcome these artefacts camera makers have fitted anti-aliasing and low pass filters [which pass low frequencies but block high ones] which simply block this fine detail. Seems crazy doesn't it? It also helps to explain the general conundrum whereby digital cameras can produce very high quality images at close quarters, but the image quality suddenly falls away as the focused distance increases, not a gradual decrease as with film. And where a digital image can suddenly appear to fall out of focus - bang, not the gentle fall away that is seen with film.

The general structured pattern of pixels on a sensor also explains why tests with line charts between digital capture and film seem so bad for digital when line frequency increases. Digital is very good at capturing irregular patterns, those that don't match the pixel structure, and very bad with those regular patterns that do. This doesn't show itself until pattern frequency matches or becomes greater than the pixel frequency. Then the digital image goes from displaying very good detail to very bad as the low pass filter kicks in and blocks the information.

Film is more even because the grains on film do not adhere to a regular pattern, it's always random. So while the overall level of resolution might not be so high, the degradation in image quality is more even. It just falls slowly away and doesn't suddenly cut off.

As sensor counts have increased some camera makers have reconsidered whether the filters are better fitted or left out. In the past Kodak DSLR's have often been fitted with removable low pass filters so the user could decide whether to use them or not, and with their last DSLR's, the 14n/c pro, they left the filters out altogether, considering that the high pixel count meant any artefacts present did not show under normal reproduction sizes, only at actual pixel size. Leaving the filters off gives far higher resolution and thus much sharper images. It is thought Nikon have taken a similar course with their D2x, perhaps not removing the filters altogether but reducing the level of their effect, as despite it having 4mp fewer pixels than Canon's 1Ds 11 - 12.8mp v 16.7mp - it is able to resolve just as much detail.

There is another reason this might be so, and another factor in sensor counts that must be taken into account. Lens resolution. The lens fitted to a camera determines how much resolution a camera will be able to give since it produces the image light that falls on the sensor. The higher the resolving power of the lens, the better the overall image quality will be. When the pixel count is increased so must the resolving power of the lenses used to make any advantage of the increase in pixel numbers worthwhile. So increasing the pixel count on an APS-C sized sensor from 6mp to say 10mp or 12mp as Nikon have done, making the pixel frequency smaller, brings with it the requirement to increase the resolutions of the lenses used. If this is not done, then a lot of the advantage of the increased pixel count is lost. 

It is for this reason that lens makers are introducing new lens with higher resolutions. It also explains why high quality prime lenses with their higher resolving powers are currently staging a comeback. They produce the levels of resolution needed, which the majority of standard 35mm film lenses cannot, especially with high pixel count sensors, another reason many makers continue to use 6mp sensors rather than higher counts.