Colour - Profiling colours from Digital Images

A New Approach

As part of this ongoing project looking into various aspects of bird ID from digital images I have arrived at the prickly question of colour sampling and profiling colours from digital images.  I would have been sceptical about this area beforehand but from my research I now believe it is actually possible to describe colours very accurately, objectively and meaningfully from digital photos.  This could be really useful for forums and hopefully for biological research generally.  

The real beauty of the method proposed below is that a colour blind observer should be able to produce the same results as an observer with perfect colour vision.  There isn't even a requirement to colour calibrate a monitor.  All the calibration is done at the camera end.

Often when there is a discussion involving the subtle colours of a bird in a digital image there is more heat than light generated as it is far too easy to dismiss an argument around a bird on the grounds that the images may not have been properly exposed or white balanced.

What are the challenges to be overcome?

(1)    Digital cameras don’t see the world of colour as we do and possibly never will.  We have to accept therefore that the colours which we describe in digital images are not the actual colours as we see them but a representation of them.  In much the same way, a sound recording is not 100% identical to the sound audible to the human ear, it is still nonetheless a useful representation of it and a properly recorded sound recording can be compared with others recorded in the same way to develop audio profiles for different species and taxa.  As the work of The Sound Approach has shown, this method is now paying dividends in terms of new species discoveries and changes to taxonomy.  Could colour profiling help improve our knowledge of different species, taxa and clines?

(2)    No two cameras are the same.  Individual cameras, even of the same brand and model will vary in terms of how colours are recorded and presented.  This can however be relatively easily (if somewhat expensively) overcome using a professional photographer’s tool called the X-rite Colorchecker passport (formerly Gretagmacbeth).  This handy tool creates camera calibrations (called DNG profiles) under different lighting conditions and effectively calibrates the camera and its images to a professional standard.  It would be nice if all camera manufacturers did this work for us automatically beforehand.  For now it remains the responsibility of the camera owner to properly calibrate their camera kit for the lighting conditions in which they use it.

(3)    Colour in birds is quite variable and subject to many environmental factors including diet, health, metabolism, fading due to strong light and and feather wear and abrasion.  There are also certain structural colours (eg. iridescence) where the colours will in fact change relative to the angle of the observer and the light source, so in the case of iridescence, there is no one true colour of an iridescent feather.  While this is all true I think most experienced observers would know that in many cases there is a high level of consistency involved in how colours are displayed by birds.  What we currently lack is a standard method (in the field at least) to study and understand where and when colour can be used for ID and when it is best to avoid getting too involved with the colours of a particular bird in a particular setting due to the natural deterioration of it's plumage.  The purpose of an objective method to study and discuss colour is to use this tool to peer into this area.  Obviously ringers (banders) and those that regularly study museum specimens will have a much greater understanding of the usefulness (or lack) of colour for ID of different species and taxa.  I certainly don’t claim expertise in this area as I neither ring birds nor regularly visit museum collections.  I am merely proposing a method here to allow field birders and photographers to start to try and make sense of colour based on images obtained in the field.  I would however love to hear from ringers and those who have had the fortune of studying a great many museum specimens (taxonomists and illustrators especially) to see what they make of this new method. 

(4)    White balance is notoriously difficult to get right.  Ambient lighting changes constantly depending on the time of day and whether conditions are sunny, cloudy or the subject is in shade.  Again there is a handy tool to overcome this and manage white balance accuracy.  It is called a White Balance or Grey Card and is simply a neutral grey patch which the photographer takes a photograph of during the course of a photograph session and then later uses it to white balance all the images captured under those lighting conditions.  The grey card and subject don’t have to be in the same scene, the card just has to be photographed at around the same time under the same lighting conditions as the subject.  Note it is important to be mindful of a bird which is in shade on a sunny day.  The grey card must be positioned in the shade also.  The colour temperature of light in a shaded area on a sunny day is not the same as that under an overcast sky though they might appear at first to be quite similar.  Clouds scatter all wavelengths of light so the light from under a cloud canopy is still basically white even in the shadows.  In the shade on a sunny day there is no scattering of white light taking place.  Sunlight is polarized.  The only light source entering the shadows is in fact the sky canopy illumination itself which is blue scattered light.  A grey card can still correct for this provided the card is positioned in the shade, preferably where the subject is/was and not out in the open where it is being illuminated by direct sunlight.

(5)    What about multiple light sources?  If a bird is under a foliage canopy, or feeding among brightly coloured flowers, or drinking at the water’s edge, or near some other highly reflective surface these nearby surfaces will all effectively become new light sources - casting colours onto the subject.  This can dramatically influence how the subject appears both to the eye and to the camera.  The classic example is trying to gauge the accurate colour of a Phylloscopus warbler or Vireo sp. under a canopy of foliage – it may be extremely difficult to do so accurately due to the green colour cast by the leaves.  Removing this unwanted colour cast can be challenging, though not impossible - see HERE.  Ideally the bird should be properly out in the open when photographed, illuminated directly by the sun or under the diffuse light of a cloudy sky.  Shade is okay too provided the grey card is used correctly for shaded circumstances.

Note: the combination of the colour checker passport and grey card provide an extremely accurate and consistent measure of hue (+/- 2 increments on the sRGB colour hue scale by my current reckoning).

(6)    Accurate camera exposure is very challenging.  In any typical series of images taken of a bird there will be some images (and areas within images) which will be over-exposed and/or under-exposed.  This obviously affects the colours captured and presented.  Hue is just one of three scales needed to measure colour accurately.  The other two, Saturation and Luminance, are both affected by exposure.  Colours de-saturate as images are under or over-exposed.  There is also of course a direct relationship between exposure and luminance.  Underexposure reduces the luminance of colours while over-exposure increases the luminance.  Luminance is also dependent on the angle of incident light relative to the observer / camera so as everyone knows when you look at a bird you don’t see a flat subject with uniform colours but rather a 3D subject with gradients of colour from highlights to mid-tones to shadows.  Gauging the mid-point – the correct luminance of a colour is difficult under those circumstances.  Lastly, as if this wasn’t already complex enough, we have Dynamic Range – the limited range a camera has to accommodate areas within a scene which have vastly different luminance values.  For instance, when you are taking a photograph indoors and a window is within the photo, the camera will either over-expose the scene outside the window and subjects in the room will appear normally exposed or visa versa - the outdoor scene will look normal, while the indoor scene will be darkened and lacking in detail.  The human eye appears to work better than any camera or video recorder in being able to quickly compensate for these changes in luminance.  But, then again we must remember that the human eye is only ever focused and observing a small area of a scene and doesn’t need to look at a subject indoors at the same instant as it observes a subject outside.  The iris adjusts dynamically to whatever the person is looking at at any given instant.  A single camera image is not a dynamic environment.  It is fixed both in space and time so is not directly comparable with how the human eye works dynamically.  In fact it has been suggested that a modern digital camera  performs nearly as well as the human eye does in terms of dynamic range in an instant of time.  Video cameras produce dynamic video but are not as quick as the human eye in adapting to sudden changes in luminance, at least not yet. 

Challenge number 6 would at first appear to be insurmountable for anyone hoping to sample colour accurately and consistently, but does it need to be a challenge at all?  How important are saturation and luminance in the context of bird ID?  Or, to put it another way.  What would happen if we decided that we could fix saturation and luminance settings and concern ourselves purely with measuring hue.  Herein lies my discovery and new technique.

I have created a Colour Profile (CP) template for Chiffchaff HERE .  How it works is the observer calibrates their camera (DNG profile) then goes into the field armed with their DSLR camera set to shoot in RAW format, plus a grey card.  The final ingredients are a cooperative Chiffchaff and plenty of patience!