Sunday, 29 March 2015

Colour - Birder's Colour Pallet - Rev. 2.0

A recent focus an colour has prompted me to tidy up some loose ends.  I developed the Birders Colour Pallet (below) at the end of a process of research into colour standards and nomenclature.  The principal behind it is that we all discuss and analyse colours in bird images but we lack standards to be able to do so objectively.  The birders pallet goes part of the way to achieving that using as simple a tool as possible.  Many of the other elements are referred to HERE.

Rev 2.0
I recently revised the pallet nomenclature slightly as I was unhappy with the naming in the region containing buff, ochre, tawny & cinnamon.   The accurate naming of colours is far from straight forward, least of all online, where no agreed nomenclature exists.  I started this process by leafing through my collection of field guides and studying the list of birds of the world for colours most often used in the naming of birds.  I then created the simple colour matrix above consisting of 24 hues, each with 3 saturation increments.  Lastly I began the task of trying to assign names to each of these 72 colour swatches.  This required an internet search for each colour and a careful analysis of the hues and saturations most often assigned to those named colours online.  What we have is a simple standard.  It may not be perfect but it is a good start.  At the end of the day, it doesn't matter what names we apply so long as we have a standard which we can reference.

Colour nomenclature altered among the blue hues in REV. 3.0.  Pallet above is current Rev 3.0.

The third scale and intermediate colours
Colour classification requires three scales.  In the diagram above we have two of those, namely hue and saturation.  The third scale is luminance, i.e. the brightness of colours.  Lighting is forever changing the appearance of colours but what changes most notably is luminance.  Look at any modern car for instance, with its carefully and stylishly crafted lines.  It has likely been painted uniformly in one colour, but the angles of each component reflect a range of tones or luminance levels as it moves through it's environment.

Therein lies the problem for any field-based colour pallet.  In the field it can be very hard to measure the accurate luminance of a colour as it would appear under optimum laboratory or museum conditions.  And yet, as humans we are masters at simplifying this complex world of light and colour.  Ask any child to name the colour of a car and they will give just one colour.  They will not be confused by lighting.  We are accustomed to the complex tones that define the colours of objects in our environment and from an early age can attribute a Colour Constancy to objects whether lit well, partially or poorly.  So, if we are to include a luminance scale in our pallet (which really we must) how can we apply some sense to it.  In the initial posting I advocated using just three luminance  increments and I still feel that is appropriate.  These are simply light, dark and a 'silent' mid-tone.

What I haven't done so far is clarify what I feel should be the line between these increments.  I think the sensible points would be Dark = 0-33% luminance, Mid-tone = 34 - 66% luminance and Light = 67 - 100%.

The other loose end which I left unclear from the illustration above is a clarification around the naming of intermediate colours.  Hopefully this is a bit clearer from the illustration below.

Why all the alternative names?  Wasn't this supposed to be a simple standard I can hear you ask?

Well, the simplicity is in the pallet of 72 swatches.  I believe in keeping the nomenclature a bit more fluid for a couple of reasons.  Firstly observers and taxonomists it seems have different naming preferences.  Cinnamon seems to be a word that North American birders might be more comfortable with.  While Tawny might be more familiar to European birders.  It doesn't matter which is used, provided everyone is applying the same pallet of 72 swatches.  The second point is that when we have a hue and saturation which sits roughly equidistant between four swatches, such as the example above, it would be very nonsensical to try and create a quadruple-barrel name.  Instead I recommend crossing diagonal swatches and allowing the user choose which name they deem more appropriate.  Once again, the chosen name is inconsequential once everyone has access to the pallet and understands how to use it.

The last point to note from this illustration is the latitude given to each swatch and intermediate.  I have tried to strike a balance where the primary swatches have more latitude than the intermediates, and I hope it works.

Colour Pallet Increments Explained
Colour Swatches = 72
Hues = 24 in increments of 10, latitude +/- 3
Saturations = 3 in increments of 80 +/- 20
Intermediate Hues = 24 in increments of 10 +/- 1
Intermediate Saturations = 2 in increments of 80 +/- 15

Wednesday, 25 March 2015

Human Bias - The Dress Viral Phenomenon

What started out as a viral phenomenon on social media in recent weeks quickly spread into the mainstream and scientific discourse.  Everyone seemed to be talking about the dress.  I first heard about it when a discussion broke out in the office where I work, then every office across the building.  Then it even made the evening news.  Every discussion was divided between those who saw a blue and black dress and those who were equally convinced the dress was white and gold.

The scientific explanation appears to centre around ones individual perception of the lighting in the image.  Some are happy to accept that the dress is reasonably well lit and that it's blue colour in particular is correct.  Others are somehow compensating for the bright sunlight and are discounting the blue colour.  What was most interesting of all was to note the reaction of people who discovered their analysis was wrong.  The dress is indeed blue and black, as confirmed by the designer.  But it seems that it is very hard to teach the brain based on this new information.  Human bias is a powerful thing as we know.  Once the brain locks on to an idea it can be difficult to change.

Sampling of the blue fabric from the unaltered original image confirms that it is clearly blue and this is how we should perceive it from a screen.  However the stats suggest that as few as one third of people actually perceive it this way.  The black material of the dress is not so clear.  That material is reflecting ambient light and doesn't look particularly dark in the image.  Black as we know is not a colour at all but rather a luminance level.  The image as a whole is underexposed and lightened so the darker tones are not well expressed.  I.e. blacks are not entirely black in the image.  In fact in this image this material measures as a dark golden or brown colour.  It is possibly this element is part of the trick.  But some of the scientific argument also points to the bright, back light as being relevant, and possibly the real trigger.  There is obviously much more to our perception of colour and light intensity than we might first suspect.  

In the image above I have used the levels tool in Adobe Elements to adjust the white point (lower left inset) and black point (lower right inset).  It would seem that a white point adjustment alone is enough to closely match the white and gold dress phenomenon, suggesting that it is primarily a lighting perception phenomenon rather than a colour phenomenon.  That said, those who observe a white dress may be using a touch of white balance correction to remove any remaining pale blue from the image as well.

It is not unusual of course for people to disagree over optical illusions.  This image is well known to most as it may be described as depicting a young woman with her neck exposed, looking away or an older woman, wrapped up, with her face in profile.  

However, with this illusion it is easy for people to appreciate both options.  With the dress this doesn't appear to be the case.

The checker shadow illusion (below) may be similar to this phenomenon but unlike that illusion which seems to fool pretty much everyone, the dress seems to divide people in the ratio of about 2:3.

The video below is some of the clearest online analysis.  14 seconds into this video a female subject appears to experience a change in her perception from the more accurate blue and black colouration to the brighter, more washed out, false white and gold colouration.  This would suggest that there is a certain, elusive trigger for it which we should presumably be able to access, as opposed to there being an inherent difference between the visual perception of two cohorts of the population   I am sure someone will figure that out eventually and feed it back to us over social media.  For now it remains an interesting party trick and source of debate.

If the back lighting is the trigger for this anomaly, this might suggest we should watch out for this surfacing in the analysis of underexposed, back lit bird images.  I am sure there will be plenty of debate about this for years to come.  As for this blog, well this lesson has reminded me that readers of this blog may well perceive the image content here in different ways.  I believe in presenting not only what I can see in the images but what can be verified using the tools discussed.  In a birders colour pallet I came up with a method to measure and describe colour without even requiring colour vision.  This is one way in which I think we can get around the subjectivity of colour analysis.

Tuesday, 24 March 2015

Birds and Light - Under Foliage Canopy

Light and shade in the forest
For many of us in the temperate zone in the Northern Hemisphere the spring equinox marks the first real signs of regrowth within the temperate forests.  Lighting under a foliage canopy is challenging at the best of times but the chlorophyll-rich, soft, new growth of the late spring and early summer presents an especially tough obstacle for observers and photographers alike.

Green Transmitted Light
This green light which bathes northern deciduous forests in early summer is put to use by many a dingy warbler during nest building and while gathering food for a hungry brood.  For the observer, this light can increase the challenge for bird identification.  This is one of the reasons most birders would probably consider forest birding the most challenging of all environments. An autumn, winter or spring tristis-type Chiffchaff is somewhat less challenging to photograph and white balance correctly owing to the relative lack of foliage.

Possible Siberian Chiffchaff Phylloscopus collybita tristis, Ireland in spring carefully colour balanced using a combination of grey card and DNG profile (more HERE).  However, faced with the same bird under a foliage canopy it is far more difficult to appreciate and photograph it's subtle buff tones.

I have already looked at this challenging light in detail HERE.

Inca Jay Cyanocorax yncas, Venezuela.  Note that, while white balance correction using a grey card does a remarkably good job, it may not be possible to remove all of the green cast, especially around the edges of our subject, where nearby foliage may increase it's local intensity.  A white balance correction is uniform across all pixels so it relies on the colour cast being completely uniform.  Which, of course, it rarely is.

Key to this diffuse light is the translucent property of leaves.  As leaves mature they become more opaque and the canopy also closes, absorbing most of the remaining light, so very little light reaches the forest floor.  This green cast is diminished and is replaced by a darker, gloomier light as the summer progresses. These are also the conditions one might associate more with tropical and evergreen forest birding.

Low Light Intensity
In rainforests there are specialists for every niche, including many that inhabit the gloomy undergrowth and floor.

Plain-backed Antpitta Grallaria haplonota, Venezuela, photographed under a very dark rainforest canopy.  Luckily it remained still and I was able to control camera shake for the 1/50th second exposure by propping the camera lens against a log.  The green haze in this instance is defocused foliage between the camera and the bird.  Without a grey card and no neutral greys in the shot, white balance in this case was by trial and error. The same goes for the image below.

Andean cock-of-the-rock Rupicola peruvianus, Venezuela.  The exposure time required for this image was 1/6th of a second.  Without a tripod it took quite a few attempts to obtain a sharp image.

Dappled Light and Other Factors
The light transmittance by leaves is also governed by the intensity of the daylight above plus the usual impacts of shadow, cloud cover and time of day.

If forest birding and photography wasn't challenging enough, perhaps the most difficult task of all is to obtain an image of the bird in the open, unaffected by shadows and obstacles.

Dappled light plays havoc with exposure and dynamic range and, in the process introduces artefacts.

To conclude, photographs taken under a foliage canopy will often present with image quality issues.  The lighting in this environment is challenging for various reasons and it can take a lot of patience to extract an identification from these kinds of images in many cases.

Wednesday, 11 March 2015

Field Marks - False Field Marks

Before I get into this in detail, a quick recap.  Bird identification may be based on a range of factors and field marks are certainly an important consideration.  We can often neatly characterise field marks as being either BOLD or BLAND based on a few characteristics.  These characteristics may include luminance, contrast relative to other surrounding features, the range of tones that go to make up the field mark (i.e. tonal range), and lastly how sharply defined the edges of a field mark happen to be (i.e. it's edge definition or sharpness).

As the analysis so far has shown, bold field marks are robust and remain detectable even in many low quality images.  The characteristics that define bland features however also tend to make them susceptible to being masked or lost as image quality deteriorates, be it through exposure, focus or white balance errors for example.

In addition to our image quality parameters, we also have to consider the overall lighting context and how light interacts with the anatomy of the bird, including translucent as well as opaque features.  In this posting I am going to broadly explore the factors that give rise to false field marks in digital images, hopefully setting the stage for the next few postings in this series.

A Trick of the Light
I have a whole page devoted to birds and light.  I intentionally started this series of postings all about field marks by looking at the interactions between light and avian anatomy.  I chose this as a starting point because lighting is such a fundamental part of this challenge.  Awkward lighting is often seen as a menace for identification but it needn't be that way.  If we have more than one image we normally have sufficient basis to rule out lighting tricks.  

The supra-loral shadow is such a consistent structural feature that is is often listed as a field mark for the identification of Booted Warbler (Iduna caligata), yet it is just in essence a trick of the light which can vanish with a slight movement of the bird's head.

The image above illustrates a few other important points.  The definition of the feathers on the crown and breast of this bird are made possible by shading.  An inexperienced observer might confuse these shadows as being fine streaking but this is one of the very first pitfalls we start to overcome as bird observers.  Shadows can also be mistaken for larger field marks.  The dark spot at the cleft where the wing meets the breast (centre image) or the cleft at the base of the throat (top left image) again might both confuse inexperienced observers.  Again, the more we familiarise ourselves with avian anatomy the less likely we are to be thrown by these features.  The two dark shadows running from the nape onto the breast are again something which an inexperienced observer might mistake for an actual plumage pattern, especially as these shadows are present in all these images.  Again, with experience comes the realisation that this pattern is not found in any species and even a simple analysis of the markings compared across two or more images will reveal that they are not located in the exact same place on the bird in any two images.  Thus, this could only be explained as shadow cast by nearby twigs.  And yet, as experienced observers we all still regularly fall foul of such tricks of the light, both in the field and in photographs.

While shadows tend to be the biggest cause of false field marks, highlights and glare can also introduce an element of confusion.  The subtle reflection of light off the base of the feathers of this Little Stint (Calidris minuta) mirror the pattern of juvenile Semipalmated Sandpiper (Calidris pusilla) to an extent.  Again, an experienced observer might pick up on this false impression based on the lighting of the primary feathers, or at the very least might seek confirmation from another photograph to determine if this is a real effect or merely lighting.

Incidentally some birders might be inclined to refer to this example as a lighting artefact.  I think this use of the word artefact is incorrect for reasons explained HERE.

Lighting and exposure are closely linked.  If lighting confuses the novice then image exposure errors give rise to even greater consternation.  Good exposure control can be key to the accurate capture of field marks in bird images, especially bland field marks.  Incorrect exposure can certainly make the challenge of identification more difficult.  But on the other hand exposure adjustment allows an observer to see beyond the dynamic range of the eye and peer more deeply into shadows and highlights.  So there may be positives to be gained even from under or overexposed images.  The key is to be able to carefully and properly interpret the evidence.

Image Lighting Tools can be used to pull more detail from JPEGs but can rapidly introduce weird and false impressions including the appearance of false field marks.  Take the heavily modified image below for example.  By using lighting adjustments it is possible to work out some additional detail from the shadows but the colours are now wildly incorrect and misleading.  Without due care the ID of this bird can take a rapid turn down the wrong path.  The ideal for image lighting adjustment is to work on maximising image content from RAW files.

Relatively low resolution may not itself introduce false field marks to an image.  There may be a temptation to try and interpret detail that isn't really there.  One of the initial motivations behind the Image Quality Tool was to try and define a point at which pixel resolution really starts to undermine image quality for identification purposes.  Extreme low resolution does introduce artefacts including pixelation, aliasing, moir√© and others and these could certainly be mistaken for real field marks.  But, at that point one would have to be pretty foolhardy to try and make definitive identification judgements where fine field marks are concerned.

While it may be highly tempting to say that this Diver (Gavia sp.) has a dark chinstrap can we really be certain or should we err on the side of caution?  At this pixel resolution I think it is just too close to call, especially were we to have just this one image from which to base our conclusions.

There are a variety of different mechanism that affect the sharpness of an image including environmental, lens and in-camera artefacts.  Each of these introduce their own particular suite of anomalies, some of which might give rise the false field marks.  I hope to delve into this in some detail to try and establish some analytical boundaries.

This image of two Magellanic Snipe (Gallinago paraguaiae) has been modified.  The bird in the background is naturally defocused by the camera lens.  The bird in the foreground on the other hand has been artificially defocused using a Gaussian blur tool.  The rear bird appears to show a double eye-ring which is not replicated by simply defocusing the forehand bird.  The double eye-ring effect could be due to motion blur or simply due to imperfect bokeh created by the aperture as light passes through the lens (for more see HERE and HERE).  Clearly, when it comes to focus we have to concede at some point that an image is simply too blurry to be analysed.  The question is, where do we draw the line?

Colour (White Balance)
Colour errors during image processing are a standard problem.  Incorrect white balance will inevitably lead to an incorrect representation of colours.  This is particularly relevant if the colour field mark is bland and if we are trying to sample and read the colour as accurately as possible.  The problem is explored in detail HERE.

Accurate colour management is essential for accurate colour rendition.  In order to be able to meaningfully compare these two Chiffchaff images for example, perhaps with other images taken with different cameras in different locations and settings, all images must be calibrated to a standard.  The standard requires the creation of a DNG profile using an X-rite colorchecker passport and then also using a grey card for white balance correction in the field.  There are also some additional considerations that might require the creation of what I call a Colour Profile (CP) for trickier subjects such as this.  For more see HERE.

Artefacts come hand in hand with reduced image quality and are most likely to be confused with real field marks where the subject is depicted in very low resolution.  The nature of most artefacts is that they are unlikely to be mistaken for a recognised field mark.  But some have characteristics that make them easily confusable with the real thing.  Such as the similarity between sharpening halos and feather fringes or moir√© and scalloping.

Wednesday, 4 March 2015

Colour - Chiffchaff Colour Profile Revisited

While researching colour in May 2014 I came up with a concept which I call Colour Profiling or CP.  The idea is to use digital colour management and colour sampling techniques to try and create colour profiles to aid in separating very similar species/taxa based on digital images.  I thought Chaffchaffs would be a good place to start and I came up with a CP template for Chiffchaff which I have been trying to test ever since.  

While I am still working on this concept it has taken me considerable time to finally catch up with a tristis-type Chiffchaff.  After a rather poor autumn for putative tristis, this winter has been remarkable here in Ireland, yet I havn't been fortunate enough to catch up with any until now.  It is quite late in the winter so plumage wear is a factor but, in any case I managed to get a nice shot of a tristis candidate at Rossmore Quarry, Co. Cork today.  This was one of three birds present, none of which were calling today (so I can't be certain of the identification of this individual).  Nonetheless this was a useful exercise as I was able to compare a collybita photographed under the same conditions.  I played a brief segment of tristis calls and was immediately buzzed at close range by the three putative tristis while a nearby collybita remained rather disinterested (for what that is worth).

Possible Siberian Chiffchaff Phylloscopus collybita tristis, Rossmore Quarry, Co. Cork.

Typical Common Chiffchaff Phylloscopus collybita collybita, Rossmore Quarry, Co. Cork.

Prior to creating a colour profile from the images above I first calibrated their colours using the X-rite Colorchecker Passport.  For more on this process see HERE.  

The CPs below were created by sampling colours from the above images.  The CP is a colour map made up of carefully selected colours samples.  I am not entirely happy with the way the CP is set up at the moment so it will need a bit of rework but hopefully the value of it is apparent from this illustration.  The colour nomenclature which I have used here is based on the Birder's Colour Pallet which I created for the blog last year.