Thursday 11 February 2016

Gestalt - Comparative Photographic Analysis (CPA)

When visiting Swedish birder Sture Persson encountered a smallish swift with a white rump at Bull Island, County Dublin on the 25th December, 2002 he must have assumed all his Christmas's had come at once!  He managed to obtain five slide images of the bird which he published together with a write-up in the journal Birding World (BW 16.1).  Images were also published online on an Irish Birding webpage which unfortunately has since been taken down.  Here is a scan of the sequence of images from Birding World.

Initial impressions seemed to fit White-rumped Swift Apus caffer, a species which breeds in small numbers in Spain and Portugal but which had never reached Ireland or Britain at that point, least of all in December when these birds should be in Africa.  I have a particular interest in this record as it prompted my first ever posting on a public internet forum, the long-running Irish Bird Network IBNListserve.  Having seen White-rumped Swift in South Africa in 2000 I queried the extent of white on the rump of the Dublin bird.  Thus began an interesting online debate around what at the time were pretty rare and instructive images provided by various contributors.  We often forget that the explosion and wealth of online bird images is still a relatively recent phenomenon.  As the years passed the record became a cold case.  During my 5 year stint on the Irish Rare Birds Committee (IRBC) I was fortunate to review and contribute to the assessment of this record but had stepped down while the review was still in progress.  As part of my analysis at the time I used a Comparative Photographic Analysis technique, using multiple images from multiple species to try and assess the shape and relative proportions of the mystery bird.

I havn't given much thought to this remarkable record again until recently.  Working with various tools and techniques for this blog I have again turned my attention back to Comparative Photographic Analysis (CPA from here) as a potential tool to help unlock a tricky identification from bird images.  In the past week I have added considerably to my original analysis of the Dublin Bay Swift and decided to make a new submission to the current IRBC in the hopes of reopening this cold case.  I wont present my conclusions here but I would like to share some of the techniques which I have applied to my analysis.

Photomontage
The human brain is believed to function by feeding new information into an already nearly fully formed picture of the world.  We do not wake up each day and approach the world afresh but instead start with an already pre-formed picture of the world as we perceive it, which we merely tweak and update here and there.  On the one hand it is clear we certainly bring a very biased view of the world to the table and it does lead in many cases to situations where we fail to notice things which are right under our noses so to speak (see HERE for example).  On the other hand this approach allows us to filter vast amounts of information quickly.  Thus we are remarkably adept at making quick comparisons, finding the similarities between objects and making quick, reasonably well-informed decisions.  Is there a way we can harness what's good about this approach to good effect here?

I often find that when I want to form an overall impression of a bird based on a collection of images, it helps to place them all in the same space.  The figures above are presented just as they were photographed.  In some the bird is facing left and in others to the right.  In some the bird appears closer and in others further away.  I started playing around with the idea that it might be worthwhile trying to put the images together in better sequence, all facing the same way and in relative scale.  Ignoring the blurriest of the five images here I have placed the remaining four together in sequence.  For simplicity I have decided to work in greyscale.


For my brain at least this is easier to assess and collate than the original montage.  How about you? 

Comparative Photographic Analysis (CPA)
The main points of discussion concerning the Dublin Bay Swift are the extent of white apparent on the rump and the possibility that Pacific Swift (Apus pacificus) or perhaps even Horus Swift (Apus horus) cannot be readily ruled out because of the limited photo set and the potential for foreshortening, defocus, low resolution and other issues to influence the analysis.  

To address all of these concerns I have drawn on a couple of techniques.  The first is only possible with thanks to the birding community who are willing to share their amazing images with the world.  I have compiled a range of these images for the purpose of research and analysis.  I apologise in advance if anyone is unhappy with this use of their image and if requested I will of course remove same from this blog.   But I hope that all birder's will see the value in what I am trying to present here as research.  Once again it is with gratitude that I thank everyone for making these images available online.

Here is a comparative analysis of the Dublin Bay Swift (bottom) alongside various images of Horus (top row), Pacific (second row) and White-rumped (third row).


For me this collection of images has been key to cracking the puzzle of the Dublin Bay Swift.  Key to creating this photo montage was the appropriate selection of a benchmark from which all other images were scaled.  In this case I used the head and body.  Swifts tend to keep their bodies fairly rigid and streamlined so it makes for a good benchmark.

Among the useful points gathered from this exercise was the realisation that only one of the Dublin Swift images was remotely in focus (2nd from left).  Next I observed that the left-hand image is the least useful for analysis because foreshortening makes it impossible to judge wing length accurately.  Note how nearly identical the White-rumped and Pacific Swifts appear when viewed from this angle.  In reality Pacific is much longer-winged than White-rumped.  That in turn drew my attention to images 2nd and 3rd of the Dublin Bay Swift and the apparent shortness of its wings.  I also noted the apparent differences in head, body and tail proportions and the slight wispy, up-turn to the outer tail feathers of the Dublin Swift.  This was all possible with thanks to the internet and those who share their images and their views online.

Simulated Artefacts
Having been working with sharp images from the net I didn't quite have the images necessary to investigate the impact of defocus and film grain or noise on the identification so I delved back into the toolbox.  Adobe Elements 12 isn't quite as masterful as Photoshop but I was able to find everything I needed to transform beautifully sharp images into images that looked like they might have been captured of a distant Swift on a dull Christmas Day.


By applying Gaussian Blur tool it is possible to nearly perfectly simulate natural camera defocus (for more see HERE).  Owing to the small size of the subject in the original slide images resolution comes into play.  We are not talking digital pixels here but the old film equivalent, grain.  Artificial greyscale noise makes a good substitute for film grain, and having applied that to the images they now more closely approach our original Dublin Bay slides.  For comparison here is the same done with the Pacific Swift images.


Once again, by using different techniques we can glean new information.  In this case I observed that the white rump was actually well within the range of White-rumped Swift and could be explained by defocus.  It was also interesting to observe how easily the wispy outer tail feathers vanished when subjected to defocus and noise.  The subtle shape of the defocused rump of the Dublin Swift compared with the defocused White-rumped and Pacific was also an interesting result.

Reconstruction
Here is another useful tool from the Photoshop stable.  When faced with a really soft image where the features are barely discernible sometimes it helps to piece together an artistic impression of the subject using transparent layers and the careful use of transformation and eraser tools.


This, the right-hand image in the Dublin Bay sequence is a typical example of a key image which could be interpreted in a variety of ways.  I have spent a lot of time pouring over this image trying to figure out where the eye and bill sat on it's blurry and grainy head.  Or, for instance whether the body and rump were in profile or banking towards, or away from the camera.  It was only when I started to overlay segments of different images that features began to emerge from the fog of this defocused, noisy image.  This image is particularly important in settling the question as to the extent of white on the rump of this bird.  If the bird's rump is tilted slightly towards the camera then we would be seeing a rump patch that is consistent with White-rumped Swift.  This is how I have depicted it here, but I could be mistaken. If we are essentially seeing the under-tail, lateral tail coverts and rump side then the extension of white so far down the side of the rump is not within the expected range and might support the alternative identification of Pacific Swift.

Conclusions
Once again it is with great thanks to all those photographers who make their images available online that we now have to opportunity to use CPA as a tool to aid tricky identifications.

Here I illustrate all of the tools discussed here in use together.  Note this is just one possible way to interpret this image.  Using other techniques such as 3D modelling the results may look quite different as discussed HERE.


There must be many more cold cases like the Dublin Bay Swift sitting in the file marked X of the archives of rare bird assessment panels across the world.  Time to dust off those files and have a go! 

Saturday 6 February 2016

Gestalt - Using Colour To Help Describe Angles

Gestalt is all about size, structure and movement.  In analytical terms at the atomic level this all comes down to an analysis of measurement and angles.  My early postings on the subject of gestalt have highlighted the problem of accurate measurement from photographs.  In bird identification we talk about proportion all the time.  However, whether for example it's Primary Projection, Bill To Eye Ratio or Leg Proportion Analysis there are limits to how accurately we can take measurements from two-dimensional photographs, as the postings in each of these links help to explain.  

In reality, as with judging proportion by eye without the aid of an accurate ruler or guide, when we discuss these ratios from photographs we are basing our analysis on subjective approximations, not on sound measurement.  Sometimes that is good enough for our needs.  While watching a bird over a period of time or analysing a bird based on numerous images, it is possible to form a fairly good and very often a very accurate impression about the relative proportion of things.  I want to apply this type of fuzzy logic in another way here.  

If you have read much of this blog you will have noticed that I like to use visual tools to draw out and present detail from an image for presentation purposes or to help clarify analytical observations and opinions.  What if there was a useful tool that would allow someone to present the approximate angles of lines within images.  When we analyse images, mentally we are often thinking about angles and perspective but when we go on to describe those things we rarely try to quantify what we mean.  Me might say for instance that the bird appears to be slightly facing the camera or that the bird's body is tilted slightly away from us in flight as it banks.  Wouldn't it be nice to have a standard tool that we could use to present our analysis of these subtle angles within an image in a way that others can appreciate more clearly and therefore discuss and critique.  Here is such a tool.

So, its basically a colour wheel or pie-chart.  When assessing angles in an image the only stipulation would be to specify that the tool is being used in a horizontal or vertical plane.  Taking the image of the Northern Wheatear Oenanthe oenanthe here is my appreciation of some of the angles in the horizontal plane.  


Someone else looking at this same Wheatear image might say, "the head looks like it is tilted away from the camera at the same angle as the wing".  That may be correct and my analysis may be wrong.  Thus the debate begins.  This form of analysis is intended to generate clarity and consensus around these subtle distinctions, or at the very least to thrash out the challenges with an identification where matters like size, proportion and lighting play a key role in the identification.