Digital Camera Sensors

There are two main sensor types currently used in digital cameras. The Charge Coupled Device (CCD), and the Complimentary Metal Oxide Semi-conductor (CMOS). Both work by converting the light hitting them into amplified electrical signals, the amount of amplification required depending on the size of the individual photo-diodes on the sensor. More commonly known as pixels, these provide the PICture ELements that make up the image a sensor supplies. 

Traditionally the CCD chip has been used in imaging devices as it produces the best quality information. Recently efforts have been made to use CMOS sensors instead as they are not only much cheaper to manufacture, they consume less power, an important factor in keeping digital camera power needs down as sensor resolution rates rise. The problem has been achieving the same quality of image using these sensors and until recently Canon were the only maker to have had a reasonable degree of success. All Canon DSLR's use CMOS sensors which they manufacture themselves.  Other makers have continued to use CCD's but the latest cameras to arrive are now increasingly using CMOS sensors as the technology improves. However, this has now had the effect of making CMOS chip manufacture actually dearer than that of CCD's, making the switch to using CMOS chips in all digital cameras, which was the initial aim, seem less likely.

Most sensors use a square checkerboard gird pattern design - matrix -  for photosite (pixel) location. Because human sight is less sensitive to the green colour spectrum, and each pixel can only record one of the three primary light colours Red/Green/Blue which are combined to produce an image, the colours are recorded in the ratio 50% green, and 25% each red and blue. As each pixel only records one colour, the information provided is interpolated to produce the final image file. This is known as Bayer interpolation. Interpolation is a method of adding information by comparing adjacent pixels. In this case the information added is colour information. It takes four pixels, two green and one each red and blue to compile the correct colour information for all four pixel locations. So a 3mp chip's file will be 9mb, a 4mp's - 12mb, 6mp's - 18mb, and so on. These file sizes are those arrived at before the file is saved as a particular format, Jpeg, Raw or Tiff. The only problem is that colour artefacts occur when the interpolation gets it wrong because it's basically guesswork, very sophisticated guesswork, but guesswork none the less.

Each pixel has colour filters over it to set the colour it records. Below are some illustrations of the way the pattern of pixels is arranged on most sensors using the Bayer interpolation pattern. Because the vast majority of digital cameras use a sensor with this pattern arrangement, the sensor has acquired the generic title Bayer, and cameras using it as Bayer sensor cameras.

          Basic Sensor Grid                                 Blue Sensor Array                            Green Sensor Array

                                      Red Sensor Array                              Combined Bayer Patten

There are several different variants of the basic sensor design made by individual companies. Perhaps the best known are Fuji's 'Super CCD' which have octagonal instead of square pixels arranged in a diamond pattern matrix, and Foveon's X3's which use three layers of pixels instead of just one.

Foveon sensor

The big problem with Bayer interpolation, as with any interpolation, is that the 'missing' information - colour in this case -  is made up by guesswork, which introduces colour artefacts.  The idea behind Foveon's sensor is that as it has three layers of pixels, each layer collecting one colours information, no interpolation is needed, so no artefacts, and image quality will be better. This is possible because Foveon realized that as silicon records light colour - wavelength - to different depths it could layer pixels to record one specific wavelength for each layer and cover the three colours. 

There is much debate, but no firm conclusion, as to whether these chips give any better overall image quality. This is because as sensor counts increase and pixels sizes become smaller these artefacts, although present, are not viewable except at actual pixel size levels. One advantage that does exist is that image sharpness is higher, and image clarity 'cleaner', especially at high magnification, i.e. pixel level.

Recent information has revealed that there are downsides to the design. Not all artefacts are actually removed, and the thinner silicon layers used lead to a reduced ISO range and performance. It also appears the design might not be 'scaleable' in the sense that the dis-advantages that exist overtake the advantages as pixel size reduces, i.e. as the number of pixels on a sensor are increased to produce higher resolution images.

                                                               Foveon three layer sensor

Fuji super CCD

Another maker using a different type of chip is Fuji, who have produced the Super CCD using octagonal pixels instead of square. Their original decision was based on the belief that they could make the pixels closer together as a result, giving either more pixels per size of sensor, or larger sized pixels per count. Going through several revisions the latest version uses two distinct types of pixels. Large ones as main collectors - S pixels, and small one's - R pixels, as secondary collectors. The first SR super CCD had two pixels located at one site. One large, one small. The current version has them separate. The small pixel sitting in the space between the large ones. Fuji's recent development has been concerned in trying to widen the dynamic range of their sensors and the ISO range.

The sensors had not been seen as making a difference with the small sized chips in digicams, but it's been a different story with regard to the their D-SLR's, the S1/S2 and new S3/S5 pro, where it does seem to result in improved images. However with the latest Fuji prosumer digicam, the S9000/9500, improvements do seem to result as the camera has a wide ISO range, 100-1600 with the lowest noise levels yet seen from a digicam. All the more remarkable as the sensor is a 1.1/8" type with 9 mp squeezed on to it.

                   Normal CCD - Square Pixels                 Fuji  Super CCD SR 11 - Hexagonal Pixels

                                  note - these illustrations are not to any scale, or in relation to each other

Although this design has been proven to provide a much wider DR (tonal) range than conventional Bayer sensors, the interpolation used to produce the image, both colour and tonal, tends to offset this advantage, (Bayer's only interpolating colour and not tonal information). This leads to an image that does not have the same resolution as a Bayer image for what is claimed is a similar pixel sensor count, Fuji counting both the small and large pixels as equal for resolution purposes, i.e. 6million 'S' plus 6million 'R' equals 12mp.

New Kodak sensor array designs (2007)

Most people have heard of and are familiar with the Bayer sensor. It's a de facto standard these days as far as most image sensor designs go. Few realize however that the sensor design is named after the Kodak scientist who invented it in 1976, Dr. Bryce E. Bayer of Eastman Kodak, and refers to the particular arrangement of color filters used in most single layer digital image sensors to create a color image. Today, almost all color image sensors are designed using the “Bayer Pattern". Now Kodak have come up with a different, some would say improved design, intended to overcome the problems associated with low light levels/high ISO and the resultant high noise levels. 

Most digital camera sensors collect colour information, and this is done by using colour filters over each pixel site. If colour filters were not fitted then all that would result would be a tonal image with no colour information, in other words black & white. This is often referred to as panchromatic. One advantage of not fitting a colour filter over a pixel site is that it becomes more sensitive to the light falling on it.

Kodak have used this basic concept to come up with the idea of a sensor grid array that uses panchromatic pixel sites as well as those fitted with colour filters to overcome the problems associated with low light levels, small pixels sites, and the problems that arise of collecting enough light to construct a viable image, and the high noise levels that result when high ISO's are employed to offset these situations.

Here are some extracts from the press release:-

 " The new approach builds upon the standard Bayer pattern by adding panchromatic pixels – pixels that are sensitive to all visible wavelengths – to the RGB pixels present on the sensor. Since no wavelengths of visible light are excluded, these panchromatic pixels allow a (black and white) image to be detected with high sensitivity. The remaining RGB pixels present on the sensor are then used to collect color information, which is combined with the information from the pan pixels to generate the final image." 

" This technology increases the overall sensitivity of the sensor, as more of the photons striking the sensor are collected and used to generate the final image. This provides an increase in the photographic speed of the sensor, which can be used to improve performance when imaging under low light, enable faster shutter speeds (to reduce motion blur when imaging moving subjects), or the design of smaller pixels (leading to higher resolutions in a given optical format) while retaining performance".

The Bayer pattern sensor array is at the bedrock of current sensor design and image performance, and has been since it's development over 30 years ago, so a new or revised design has many implications for the future of digital capture. We doubt Kodak would make an announcement like this without a certain level of knowledge of the likely impact such a design will have, in other words how well it's works. And if the sample images/comparison shots supplied are anything to go by, then the design works very well and is sure to be implemented in many digital cameras in the future.

No one solution is usually ever an answer in itself however, and this won't be the only answer to noisy images taken in low light levels with high ISO. So it won't replace, at a stoke, all the anti-shake systems that have come into being. But it is an alternative, and a good one at that, for the many situations that occur when a fast shutter speed is needed, not to overcome camera shake, but to stop motion blur. And that can happen in all kinds of light levels, not just the very low ones.

However this design concerns more than just high ISO noise levels. It works at such a basic level that it has image quality implications right across the board for all digital cameras of all types. If it has improved image quality implications for the small digicam sensors then this applies equally to the larger sensors used in DSLR's. Indeed it may prove a turning point in that the difference in pixel site size is reduced to a level that has far less significance than it currently does in respect of image quality. 

And that it turn might mean that the smaller APS-C size sensor designs common in most DSLR's remain the size used in the vast majority of cameras, and that those DSLR's that use full frame (35mm size) sensors continue to be in the minority.

Sensor Format

Sensors are produced in one of two formats or aspect ratios. These are the 4x3 ratio most TV's use, and the 3x2 ratio which is the 35mm film format. The former is usually found in digicams and the latter in D-SLR's.

Sensor Size.

There are several sizes of sensor used and their physical dimensions are given in the table below along with the aspect ratio. You will note that some designations used do not correspond to their actual size. Its very confusing and they should all have been replaced by better notations, but to date haven't. These have their origins in the early days of Television cameras.

You will see from the table that all the sensors are quite a bit smaller than you might expect. You could fit no less than 22 of the most commonly used 1.1/8" type into the area covered by a 35mm negative.

To illustrate their size relative to each other here is a comparison chart using different colours for each size. 

Digital Camera Sensor sizes

Legend

You can see that there are three clearly defined 'groups'. There's the full frame 35mm sensors, the more common APS-C sensors, and finally the 'digicam' sensors. You will note that there is quite some difference in size between a full size 35mm frame and the general DSLR sizes, and another big difference between these and the smallest size group, the digicam sensors. In the groups as well there is a marked difference between the biggest and smallest. It is also noticeable that the 4/3rds sensor, whilst about 4 times the size of a 2/3" sensor is the smallest D-SLR sensor by a considerable degree. 

Another big difference is the area covered by the standard 35mm frame. It is 2.3 times bigger than the largest APS-C sensor size. And 3.5 times the size of the 4/3rds size sensor.

All these differences are significant as we will see later when we come to sensor resolution and actual pixel size.

If your intrigued as to what a sensor actually looks like, here are two views of the APS-C sized CCD sensor inside a Pentax *ist-D. On the left you can see it in its surroundings. On the right is a close up of it with the size of the largest current digicam sensor size, 2/3", superimposed as a black rectangle to further illustrate the large size difference between them. The red dotted line represents the actual area of the sensor that collects the 6.3million pixels worth of data it supplies.

And just to illustrate how small the little digicam sensors are, here is a view of one in it's camera assembly, with a drawing pin besides it for size comparison. This is out of a 3mp Sony Cybershot which failed and which it wasn't feasible to repair, the lens assembly having stripped it's nylon gear train. Nothing lasts forever. The actual sensor area is of course within the square blue outline that you can just see in the centre of the sensor assembly. Tiny isn't it?