Saturday 12 September 2015

Human Bias - Camera Versus Human Eye (Part 2)

Imaging Regimes
We are all probably acutely aware of the difference between night vision and normal or day vision.  Night vision utilises scotopic vision, involving the rod cells of the retina.  These cells surround the focal point of the eye - the fovea.  The image produced is faint bluish and monochromatic.  It is also not as sharp in appearance as the image produced during the day (photopic vision) due to lower resolution and a low signal to noise ratio (SNR).

We are accustomed to obtained a sharp image by focusing on a very narrow line of sight.  This sharp photopic image is formed by the densely packed cone cells of the fovea.  In the night sky, when we attempt to resolve fine detail in a dim object like a galaxy or comet, the object often appears to vanish before our eyes.  This happens because the cone cells of the fovea are not sensitive enough to such a low level of light.  Also photopic vision may be temporarily suspended during scotopic vision.  Simply by focusing to the side of the dim object, it suddenly reappears, though somewhat blurry.  This is not an illusion.  It happens because the rod cells once again are able to register the dim light of the object shining on them, albeit it at a lower sharpness than we might like.  For a novice astronomer having to observe faint objects in this way may be disconcerting.  Unfortunately there is always a trade-off in very low light between brightness and sharpness in the visual system.

Rod cells are not completely switched off by day.  They form an integral part of our peripheral vision.  Scotopic and photopic vision can operate together on a moonlit night and towards dawn or after dusk.  It also occurs when we drive a car at night.  In our modern world, polluted by light, some of us possibly never get to experience true scotopic vision.  The mixing of the two vision systems is referred to as mesopic vision and explains why for example we can see sharply in colour while driving at night.  The different vision regimes make for an important distinction between human vision and cameras.  For a start these human adaptations dramatically expand the dynamic range of the human visual system when compared with digital media.

However camera image exposure allows cameras to make up some lost ground.  Long exposure times allow cameras to peer far more deeply into the gloom of a dark night than the human eye ever could.  I recently watched a documentary where it was reported that by progressively exposing a single image of a small portion of the night sky, hundreds of times over countless hours and successive nights with the aid of the orbiting Hubble Space Telescope it has been possible to photograph some of the dimmest objects in the universe.  Therein lies the miracle of modern photography!  At a slightly more mundane level, adjusting ISO allows us to amplify our own digital images to extract more detail and colour, though we also increase noise in the process.

Despite the eye's broad dynamic range we also struggle to perceive detail at high light intensity.  Once again, by reducing exposure time, aperture and ISO we can record digital images at extremely high luminance levels that would otherwise seriously damage our eyes.  Using such techniques we can even photograph and capture detail on the surface of the sun.

Spectral Sensitivities
Human vision operates roughly between wavelengths 400nm and 700nm.  Our night vision rods create a monotone image in the blue end of the spectrum at approximately 500nm.  During the day, while using photopic vision our cone cells are dominant.  Our cones are distributed within the fovea at a ratio of green:blue:red of 2:1:1 and our eyes are most sensitive to the green portion of the light spectrum.  But we also have a strong sensitivity to blue and red.  We cannot see beyond the violet (ultraviolet) or red (infrared) wavelengths and hence these colours define the boundary of the visual spectrum.

Typical digital camera sensors or the other hand have a slightly different spectral sensitivity and this varies from one camera make and model to another.  Digital sensors typically have a higher sensitivity to blue but with image processing this is not noticeable.  Camera manufacturers apply a ratio of 2:1:1 in terms of green, blue and red filters respectively to the surface of digital sensors to match the sensory configuration of the human fovea.  Unlike the human visual system digital camera sensors are sensitive to both near-ultraviolet and infrared.  Special UV and IR filters are used to filter out this normally unwanted light.  Birds see in UV and we can explore this hidden world in monochrome using a modified digital camera.  For more see HERE.

Gamma Correction
Our eyes do not perceive light intensity the same way cameras do.  Camera's record light intensity linearly.  Twice the light intensity equals twice the recorded signal.  On the other hand our eyes record light non-linearly.  We can for example distinguish between dark tones with a fine degree of discernment.  In other words we can perceive a contrast between very slight changes in light intensity within shadows.  However we are poor at distinguishing such relatively small contrasts within higher levels of luminance.  There is an evolutionary advantage to this trait.  It has allowed us to function well in low level light while at the same time providing our vision with enough dynamic range to function adequately even under the brightest of daytime conditions.  What this means in terms of imaging systems is that it makes sense to preserve plenty of mid-tone and shadow detail but we can afford to preserve less within the highlights.  All imaging equipment corrects for this non-linearity in human vision and this correction is referred to as gamma correction or gamma compression.

Technological Advances
As illustrated above we can, in any given image exceed either the low light or the high light range of human vision, but we can't do both at the same time. Current digital cameras don't even come close to matching the vast dynamic range of the human visual system within a single image,  Also, despite the fact that many cameras are capable of recording 16-bit images, the internet and virtually all screens and printers can only output images in 8-bit.  So there is always a trade-off and loss of some colour and tonal detail.  But technology is advancing all the time and it is quite reasonable to expect that within 10 or 20 years digital camera technology will be able to match and surpass the imaging capabilities of human sight.  In time output devices will also advance and it should eventually be possible to display an image with the same brightness, colour and clarity of human sight.  When that day comes we will probably once again begin to take for granted some of the wonderful complexities of light and imaging systems.  But for now, if we are to make the most of our camera equipment we need to understand it's limitations and the workarounds required to record what we see, and beyond.

No comments:

Post a Comment