PROJECT: NOISE

“One of the advances in sensor technology in recent years, along with increased resolution, is the massive improvement made in combating noise” (Freeman 2011 p.58).

Whilst looking at the Project Linear Capture earlier in this module, we covered the fundamental workings of a camera’s sensor and its relationship to the human eye.  Before we take the plunge to look at noise in its entirety, I think it is prudent to build our knowledge surrounding sensors a little further.

Sensor Types

The sensor is the heart and soul of our camera, feeding and serving almost every function our camera undertakes, but its main function is to capture and convert light into electronic signals, and process these into the images we make every day.  As we saw previously, these processes are highly complicated, and as technology continues to evolve, so will our camera’s sensors.

Currently, there are two sensors commonly found in digital cameras these being the CCD Sensor (The Charge Coupled Device) and the CMOS Sensor (The Complementary Metal-Oxide Semiconductor).  The fundamental difference between these two pieces of technology is that the CMSO sensor converts charge to voltage at each photoelement, whereas this conversion is carried out in a common output amplifier within the CCD sensor.  During the early development stages of sensor technology, CCD sensors had image quality advantages, which made them a popular choice with camera manufacturing companies, so the CMOS sensor was overlooked, but as technology advanced, significant changes have been made and now the CMOS sensor seems to have the upper hand.

CMOS vs. CCD

Following is a list compiled from Freeman 2011 p.30 and Langford et al 2008 p.110 outlining the pros and cons of these two sensors:

Advantages:

  1. Although CCD’s are inherently less noisy, with larger sensors, such as APS-C and full-frame capabilities, these are largely negated benefits. This is due to the fact that; CMOS sensors have bigger photosites that can collect light more efficiently; they have a reduced density of photosites, and improved micorlens and circuitry design.
  2. CCD sensors still hold superiority over light sensitivity, dynamic range (their range of tonal values is greater) and noise performance.
  3. CCD sensors base themselves on a more mature technology.
  4. CMSO sensors use much less power.
  5. It is possible to manufacture CMOS sensors in existing computer chip foundries, making them a less expensive option to manufacture, particularly at the larger sensor size.
  6. CMOS sensors optimise images by incorporating other non-light collecting circuitry.
  7. The CMOS sensor contains higher levels of system component integration on their chip (SOC – system on chip), which can simplify the camera electronics, especially in analogue-to-digital conversion and image processing.

Disadvantages:

  1. Manufacturing CCD’s is complex and there is no integration ‘on-chip’
  2. The image quality of the CMOS sensor is not as good as the CCD; this is mainly due to higher noise levels and the higher level of on-chip electronics used.
  3. CMOS has a lower fill factor and quantum efficiency

So, from the above we can see that each of the most common camera sensors have both good points and bad, and our choice of sensors really comes down to the type of camera we use, although looking at the most common DSLR models, CMOS sensors seem to be the current norm.

Noise

Now that we have rounded up our research on camera sensors, although I seem to have done this back-to-front, it is time to look at noise.

Even thought technology has come a long way, we still find some limitations and constraints surrounding the design of a cameras sensor and these can give rise to certain image artefacts, which in turn can affect the quality of our images.  These artefacts include blooming, clipping (as seen in previous posts), aliasing and moiré patterns.  Moiré patterns are the strange stripes or even bands that are produced when the level of texture detail in certain parts of a scene are high.

An Example of Moire Patterns

An Example of Moire Patterns

Probably the most documented and discussed image artefact is noise, which actually has nothing to do with the scene we are photographing, but it is a recorded error of an apparently random variation in pixel values.

The DSLR’s on the market today have seen massive improvement in their capability of dealing with noise as they not only have usable ISO settings of around 12,000, but also built-in noise reductions options (if your camera has this capability, settings can be found in the options menu).

On the surface, noise resembles the graininess found in film and when compared can be put down to the relationship between sensitivity/film speed vs. noise – but this is where the similarities end.

This photo, taken at Lau Pa Sat in Singapore has an ISO setting of 1,000, which is great for capturing the light and the image, but as you can see in the red box, there is noise present in the darker crevices of the image.

An Example of Noise

An Example of Noise

Luckily, in this digital age, noise in our photographs can be addressed, but by no means is this easy or the results always perfect.  There are in fact many forms of noise and it should firstly be defined before it can be treated.  Technically, noise is the error introduced into the image by electrostatic charges, which scientifically revolve around the signal-to-noise ratio; noise is detail, but mostly unwanted detail, which can in fact be a subjective term, as it will always boil down to preference and the design elements of our images to understand how much noise should be allowed in or tolerated.

The production of noise is very dependent on a number of variables and can therefore vary from shot to shot and camera setting to camera setting.  In most cases you can expect to come across this artefact when using high ISO settings (as above), long exposures and in images that contain large areas of smooth, undetailed shadow, which we discovered when looking a linear capture.  We can see that there are several contributing factors to the appearance of noise in images, but the fundamental cause is the lack of light photons striking the sensors receptors, which in turn causes this sampling error.

Freeman (2011 p.58) notes that there are five origins of noise:

Photon noise: caused by the way in which light falls on the sensor.  This type of noise appears as bright, dark or coloured specks and is most apparent in plain areas and least apparent in highlight.

Readout noise: caused by the way a sensor reacts and more specifically the way it reacts to the signals processed by our camera.  Also known as amp or bias noise, as it is similar to electronic noise found in recorded music and is generated by the processor, which becomes more amplified the more prominent (or sensitive) it becomes, thus by increasing the ISO the ‘amplifier’ creates more noise.

Random noise: this noise is caused by unpredictable, but inevitable differences in the timing and behaviour of electrical components.

Dark noise: caused by imperfections found in the cameras sensor.  Can increase in proportion to exposure time and temperature variations, and is independent of the image, so may be reduced by in-camera processing.

Reset noise: once a shot it taken, our camera resets the sensor to a zero position, and gets ready for the next exposure.  There may be slight differences in the timing of this, which will generate noise.

Noise can appear in different shapes and forms, and in some instances may not actually be noise at all!  When enlarging an image, luminance noise will appear as a variation in brightness, which can take on the effect of a monochrome, grainy pattern.  When recording chrominance noise, we find prominent variations of the images’ hue, and a hot pixel appears as a bright spot that is ‘stuck’ within the body of the scene, this can be visually disturbing and very noticeable.

However, there can be times when real pixel details are mistaken for image noise and it is not until you enlarge sections of your image that you find the pixels that constitute to real data.  There can also be times that the only distinction between the two (noise vs. pixels) is knowledge of the scene and the better judgment of the photographer, who needs to decide which detail is classed as noise and which is classed as real.

“Look at the image ‘Grey Texture’, you will see that there are different kinds of texture in the garment – folds, vertical ribbing and a mottling in the grey cloth.  Think about the mottling.  Is this noise or is it part of the pattern?  It could be either, but in fact, it is real detail.  Distinguishing detail from noise is ultimately subjective”

Grey Texture

Grey Texture

This is quiet a subjective question as when looking at this image, you would automatically assume that the mottling was noise or more specifically Photon noise, due to the colour specks present in the grey of the suite.  The statement above, taken from our course material states that this mottled detail is real, but the photographer has the unfair advantage of seeing the image as a whole, whereas we only have this small section to study.  If this is in fact real, I think that the detail lost in the folds and shadows is down to noise, so it may not be present where we think it should be, but it is there!

“Now look at the image ‘Turkish Dance’, this is quiet a noisy image because it was photographed at ISO800.  The noise dominates in the shadow areas, but you can see that at this level of magnification they are on a par with real detail, in particular the silver brocade on the dancers’ red dresses”

Again, the photographer has the advantage of knowing the scene and knowing what to look for during post processing.

Turkish Dance

Turkish Dance

When comparing these two images, it is very difficult to distinguish between noise and detail as both have the same effect in their darkest areas, and as the black jacket is velvet it can be questioned whether the noise seen is actually noise and not detail of the pattern.  However, I do agree that the silver brocade looks very fuzzy, which could be put down to high levels of noise in the shot.

A difficult subject to broach and something that is very subjective.  If you are just the viewer of a photograph and not the photographer it could be difficult in some situations to determine what is noise and what is pixelated detail.  The only way to get around this is to look at the bigger picture first to understand what elements are present, and then make a judgement call on the noise levels, as I think looking at only snapshots of an image can be misleading.

Source:

Reference:

Image:

Karmeba.  (2011) Moiré Pattern [Online Image].  Available at: <http://carmeba.wordpress.com/2011/10/09/task-1-images-sourcing-digitising-and-defects-2/&gt; [Accessed 21 May 2013].

Books:

Freeman, M.  (2011) The Digital SLR Handbook.  Revised 3rd Edition.  East Sussex: The Ilex Press Limited.

Langford, M., et al.  (2008) Langford’s Advanced Photography.  7th Edition.  Oxford; London.

Bibliography:

Langford, M., et al.  (2008) Langford’s Advanced Photography.  7th Edition.  Oxford; London.

Advertisements
This entry was posted in Digital Image Qualities ~ Exercises and tagged , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s