April 2018 - Digital Group - Image Noise

Meeting Notes March 2009 to 2018.
Post Reply
Posts: 148
Joined: Tue Mar 25, 2008 7:04 pm

April 2018 - Digital Group - Image Noise

Post by spb » Sun Apr 01, 2018 6:14 pm

The Google Images search has been subject to a deal with Getty whereby it may be harder to download high resolution images directly, or at least without visiting the source website. This is due to copyright infringements. This deal does not affect the many other sources of images such as DuckDuckGo.

In another related case, a New York judge has ruled that any photo embedded in online content such as Twitter or Facebook could potentially infringe copyright. Watch this space.

A blockchain crypto-art digital image of a rose has been sold to ten investors for a total of $1M. This is the same technology as used in Bitcoin. The ten rose tokens can be sold-on or gifted.

The recent helicopter crash in the NY East River was a doors-off photographic flight. It seems that the robust harnesses could only be removed in an emergency using a tethered knife. The helicopter made a controlled landing on the water but then it turned over. Nobody managed to escape except the pilot. The main purpose of doors-off flights seems to be ‘foot selfies’ in which you take an aerial shot including your dangling feet.

Like all camera makers, Canon and Nikon are suffering year-on-year sales slumps caused by smartphone cameras. These may not be great cameras but they are good enough for most people and instantly connected to fun apps and social media. Both manufacturers are rumoured to be working hard on full-frame mirrorless cameras to be released in the next 12 months. This makes a lot of sense in terms of enabling rapid innovation but raises the question of backwards compatibility with their legacy lenses, something that they both hold dear.

As reported previously, a Windows update resulted in the temporary loss of all of my USB ports. Fault-finding this involved repeatedly plugging and unplugging external HDDs which had apparently not been recognised. Unfortunately this was sufficient to kill the SATA/USB interface circuitry in two of my backup drives. A subsequent Windows update resulted in temporary loss of the ‘Safe to Eject’ notifications and another USB HDD backup drive was killed. I was able to recover the first two by replacing the external enclosures but the third is terminal as the HDD is hard-wired. This is another reminder that one backup is never enough.

We don’t think much about data obsolescence but I am pretty certain that everything lovingly stored on magnetic hard disks today will be unreadable within a generation. This is a thorny problem. An ironic example is the 1986 project by the BBC to create a modern Domesday Book to mark the 900th anniversary of the original. Contributions from a million volunteers, 147,000 text articles and 23,000 amateur photos were stored on cutting-edge Laser Disc. Unfortunately the format didn’t catch on and was expensive for schools. Within a few years the discs were unreadable to all but a few enthusiasts and attempts began to emulate and eventually reverse-engineer the text and images onto the web. It’s so ironic that the original Domesday books are as readable as the day they were written 900 years ago but within 15 years the new Domesday Project had to be rescued! Read more here. It’s not just storage technology that becomes obsolete, file formats do too. One of the problems with the Domesday project was that the images pre-dated the JPG definition and were encoded in an analogue uncompressed form. Apparently ‘experts’ are pinning their hopes on PDF/A as a self-contained archival document format which contains both the look of the document and the extractable content. This becomes vital when considering records that must survive in the long term such as property deeds, pension details or clinical trials and research. When combined with the sheer volume data stored today this is all truly scary and how many images of current times will survive for future generations?

Noise in this context is mostly caused by electrical randomness when measuring the level and colour of light at each pixel site of a camera sensor. It appears as light and/or coloured speckles across the entire image but especially visible in dark areas.

Controlling Noise. The sensitivity of a digital camera to light is controlled by setting an ISO value. As the amount of illumination of the subject decreases, turning up the ISO value of the camera can maintain an adequately exposed image, at the cost of increased noise and decreased image quality. It’s a valuable exercise with a new camera to determine the ISO level above which you find the noise unacceptable. For most cameras 1000 ISO is fine and many modern camera will happily go beyond 10,000 but it may not be a good idea to go there on a regular basis. The alternative to high ISO settings is to add more light to the subject eg with flash, maybe use a tripod and a longer shutter speed and/or a larger aperture. Auto ISO settings often have settable limits for maximum ISO which may intelligently take into account the lens, stabilisation, aperture and shutter speed.

Very long exposures of many seconds or minutes can heat up the sensor and increase noise. Many cameras offer a long exposure (and maybe also High ISO) noise reduction option which can be very effective but may double the exposure time. This could be a problem where speed is of the essence eg star trails.

Sensors. These are continually improving and although noise reduction is one objective, it conflicts with the more pressing demands from customers for size reduction and more pixels. As the number of pixels increases and the sensor size decreases, the size of each pixel receptor shrinks and noise increases. A smartphone may have a pixel site of 1 micron wereas a full-frame DSLR may have a sensor site of 8 microns. Bigger cameras have bigger sensors and generally, better quality pixels. The size of a pixel site in microns is a better measure of quality than the number of pixels. Arguably virtually all cameras have far more pixels than necessary. An A3 print only needs about 8Mpx including a margin for cropping. Sensor technology has a few tricks to reduce noise. Backside Illumination is the new buzz-phrase. This places the connections to each pixel one the back of the sensor. As a result more light can fall on each pixel and so reduce noise.

Before buying a camera it’s interesting to compare image quality and noise using the dpReview Studio Shot Comparison Tool. This enables different cameras to be compared side by side under identical lighting conditions and settings. It can be very revealing.

Post Processing. Unless you are prepared to pay for, and lug around, a large DSLR, a certain amount of noise is inevitable. So what can be done to minimise it in software? Firstly if you shoot JPGs then most noise processing is taken care of in the camera, although more can be applied. On the other hand, RAW images may receive only a little (or no) noise processing in-camera. Mostly it’s down to you. In Photoshop/Elements Adobe Camera Raw (or Lightroom) the Detail tab offers sliders for Colour and Luminance noise reduction. The former is excellent, the latter is a balancing act against loss of detail. However, with a little practice the results can be excellent unless you have pushed the ISO setting to the limit. Don’t forget that JPGs can be passed through the raw converter in exactly the same way to improve the in-camera noise reduction still further.

It’s too easy to zoom-in and pixel-peep. We can get too obsessed with low levels of remaining noise that are far less than we were happy to accept as film grain. Film Grain was an inherent property of the film making process involving silver halide crystals. Once again, more light-sensitive films had larger film grain but this was often considered quite ‘arty’ and desirable. I used to buy 1000 ASA film for it’s grain and pastel colours. It’s true that film grain was more irregular in size than digital noise but it was huge in comparison to the fine noise we see today. If we want to add simulated film grain to our digital images there are numerous filters to do this. Built-into Photoshop/Elements are a couple of basic options Filter>Noise>Add Noise>Gaussian or alternatively Filter>Texture>Grain. More sophisticated old film simulations are available elsewhere in plug-ins to Lightroom and Photoshop/Elements such as Nik, DxO, Alien Skin and many others.

Sadly film grain used to be warmly embraced as part of the creative process but today it’s more likely to be seen as a fault or a gimmicky effect together with light leaks and processing faults.

Cheers, Steve

Post Reply