Friday 22 April 2016

Gestalt - Gestalt Keys (Primary Projection)

For those new to gestalt and gestalt keys please read the introductory post (HERE) and the essential principles of gestalt keys (HERE) before proceeding.

Primary Projection (PP), or Primary Extension is one of the tougher measurement challenges.  All the structures that influence PP are fluid.  I have discussed the limitations of primary projection as a measurement tool in some detail HERE.  But despite those limitations it is still possible to measure PP accurately and consistently within certain, tight constraints.

A Question of Axes
If we accept that primary projection only makes sense as a measurement on an orthographic projection we are already part of the way to identifying a proper method for measuring PP from digital photos.  As two-dimensional projections go digital photographs may not be perfectly orthographic but they often approach it closely.  For the first problem...how do we make our bird sit in perfect side profile along our measurement axis - the X axis?

X Axis Control
We actually have a very useful and reasonably reliable way of judging if a bird is sitting in perfect side profile.  The answer is contained in a bird's wings and tail.

This little exercise reveals how primary projection should be approached from a digital image.  By viewing a bird from a slight elevation (or with it's body tilted towards the camera) to just about reveal the top of the far wing, we have a ready made formula not just for controlling the X axis, but also for gaining a perfect view of all the salient features.  Okay so this is not always 100% reliable.  A bird can hold it's wings slightly askew relative to one another.  Or the tail may have a jagged end to it.  We do however have a number of points of reference here to help make a sound judgement, including the tips of each of the tertials, primaries and tail feathers.

It is also true to say that by shifting elevation, effectively raising the camera above zero mark on the Y axis, we are introducing an angle of perspective.  This will introduce some foreshortening on the Y and Z axes, throwing our measurements out slightly in both.  However, in this case we are only interested in the preservation of orthographic (or parallel) measurements in the X axis.  PP measurements should be unaffected.

As a rough stab at PP measurement we have probably done enough already, simply by confirming X axis alignment.  Provided the structures of the wing are aligned reasonably naturally it should be possible to take a pretty good PP measurement.  But if we want a more solid measurement, plus perhaps some additional confirmatory clues, I recommend going a step further. 

Y and Z Axis Control
This is where things get tricky and we start to look towards a Gestalt Key solution.  While we may have cracked the X axis, locking down the bird's body and overall wing orientation.  We haven't yet dealt with the position of individual feather tracts of the wing.  An individual bird may hold its tertials slightly drooped or in varied alignment.  Secondaries and primaries may be clumped or fanned in a variety of positions.  To the eye, looking at a bird in side profile (even slightly elevated) it may be difficult to tell if the primaries are resting on the side of the rump or crossed over the top.  All these points have a significant bearing on PP.

I started approaching this piece of the puzzle by assessing the relative position of the joints of the wing in the hopes of finding a few suitable surface points to lock onto.  Unfortunately this seems to have led to a dead end.  There doesn't seem to be any way of accurately linking the position of the bones and joints of the wing to surface features.  Time for a simpler approach.  Having established that the far wing represents the locus for the X Axis, we only really need to be certain of the position of the base and tip of the tertials and assert some control over the primary tips and we should have our solution.

Conclusions
Accurate PP measurement is undoubtedly tricky but not impossible.  The wing itself plays a useful role in confirming that a bird has been photographed in side profile.  After that its a matter of constraining the wing structures.  That's where the concept of a Gestalt Key comes into play.  Having an overlay map as it were we can introduce useful additional morphological landmarks such as the approximate position of emarginations on the various primaries.  No tool of course is ever 100% fool-proof.  Feather wear, misalignment, moult, including growing or dropped feathers and individual variation all have a say.  But I hope that tools like this give a new perspective on an old subject.  And who knows there might even be a few new tricks and discoveries to be made with the help of Gestalt Keys.

Saturday 16 April 2016

Gestalt - Gestalt Keys (The Principles)

For those new to gestalt and gestalt keys please read the introductory post HERE before proceeding.

A Secure Solution
By now hopefully you will have had time to digest this concept which I have been developing and you can see it's purpose and potential.  The principal by which gestalt keys work is much the same as the pin tumbler lock and key.  In order for a door key to work the little serrated hollows or divots on the edge of key lift tiny key pins in the lock mechanism as the key is inserted.  Provided these pins all then align along the shear line the key can be turned to open the latch and unlock the door.  The security of any lock depends on the quality and sophistication of it's design.  I am hoping to apply these same principals to the design of effective gestalt keys.

At the moment birders already use a range of gestalt measurement tools to aid an identification.  While these tools no doubt are rooted in sound logic they all tend to lack a secure means for ensuring they are applied correctly at all times.  If we take any one of these tools, e.g. Primary Projection (PP), and analyse it's limitations as a measurement tool (as I have done HERE) the lack of rigor applied surely equates in security terms to having no lock on a front door.  The correct application of PP becomes a matter for trial and error and debate, and rightly so.

Orientation And Perspective
In the posting on lighting and perspective (HERE) I outlined one of the important distinctions between human vision and the camera.  The world as we perceive it is actually a distorted reality.  Our brain assimilates the slightly different perspectives of both our eyes to create one image and one perspective.  Not only does the brain have to balance two slightly different viewing angles but it also plays around a bit with normal perspective lines in the process.  The result is an image which never quite matches reality.  Of course the camera image isn't totally accurate either.  As light passes through a camera lens it is distorted in order to focus detail onto the small, flat surface of the sensor. This is especially true of smaller focal length and fish-eye lenses, which attempt to gather a much wider angle of view than telephoto lenses.
So how should we best present our subject for gestalt measurement?  In order to preserve scale and therefore accurate measurement we must minimize image distortion.  We therefore must design Gestalt Keys so that they approach perfect axial orthographic (or parallel) projections of our subjects.  Generally, a perfect axial side profile is what we are after.

Camera images will always contain some element of both perspective and lens distortion.  Images of birds captured at long range with a long telephoto lens help minimize perspective distortion.  Furthermore, modern DSLR cameras typically carry out image processing to correct for known lens distortion.  So, many images do approach an orthographic projection in their preservation of scale.

Digiscoping continues to be very popular, particularly as phone cameras start to produce images on a par with compact digiscoping cameras.  While the image produced by the scope or binoculars may be reasonably orthographic, thanks to very long focal lengths and high quality lens components.  It is worth bearing in mind that the short focal length of the compact camera or phone camera may give rise to lens distortion of the image beyond that point.  For more on perspective and image distortion see for instance the postings HERE and HERE.

One would hope that all gestalt tools currently in common usage are originally based on sound principles requiring preserving scale, using a proper axial profile and taking account of a bird's natural posture.  On the other hand, it can be extremely difficult to gauge a perfect side profile on a bird be it in the field or even in the hand.  In some cases, a tool might have been developed based on a small sample size.  Or, perhaps some flawed logic was applied.  For instance, if measurements were taken by manipulating feathers to align perfectly along a straight ruler, in a manner that doesn't match how these features align naturally in the field, then such measurements clearly would not be readily applicable to ID of a bird observed (or from images captured) in the field.

'Alternative' Gestalt Keys
Having already established that gestalt measurements tend to have been made based on perfect side orthographic projections this does not exclude the possibility of generating 'alternative' gestalt keys from different perspectives.  It is always tempting to try and gauge a gestalt measurement such as primary projection from other viewing angles.  We may not always be fortunate enough to capture an image in side profile.  For instance, in the images below the angle of view provides for a reasonably good view of all the relevant structures but clearly foreshortening is at play so the standard primary projection measurement will not apply.

The advantage of gestalt keys is that they apply rigid principles so that a measurement can be replicated consistently across many images.  One could develop multiple gestalt keys for the same identification problem but using different perspectives such as side, front, rear, plan and any angle in between.  Accuracy and reliability comes down to an appropriate selection of loci.  The loci used for locking down a rear profile view will differ from those used for a side view.  Needless to say the actual measurements from alternative gestalt keys will differ from the original, standard side profile key.

Hard Versus Soft Loci
Bareparts, including structures of the bill and legs have the potential to make for the best loci.  Obviously given that these are reasonably solid structures, with minimal moving parts there is a good chance of using them to lock down image orientation.  This is especially true of the bill as it is typically aligned with perfect symmetry along the Y axis of the head, i.e. it's side profile.  So when it comes to the features of the head the bill is certainly a 'go to' locus.  Having said that, we may require the resolution of some fairly minute structures of the bill to ensure that the bill is actually axially aligned.  A combination of the bill width at the base, nares (nostril) and cutting edge of the bill should normally suffice to align a bill correctly.  In the cormorant example above we don't have the benefit of a nares (as Cormorant's don't have one).  Instead the culminicorn, the bony plate encasing the culmen on top of the bill, provides an additional locus to lock down any lateral rotation of the bill and head (around the X axis).  While the overall proportions of the various structures should hopefully take care of longitudinal rotation (around the Y axis).  Of course we have intra-specific variation to consider.  So, once again development of an effective key will require some trial and error.

Soft loci are far more challenging.  In future posts I will be tackling primary projection among other challenges.  For many gestalt tools we compare the relative position of feather tips.  Clearly feathers and the tracts they lie in are highly dynamic.  Determining appropriate loci for these soft structures will be a challenge but I am confident solutions can be found.  I expect it will take a good deal of experimentation, trial and error to find the right matches.

Sunday 10 April 2016

Gestalt - Gestalt Keys (An Introduction)

So far all my analysis has concluded that gestalt is not accurately measurable from digital images using the classic techniques birders have been using up until now.  If only there was a key that could unlock the tricky subject of gestalt.  



The Problem
Gestalt is complex.  Take this Bulwer's Petrel (Bulweria bulwerii) for example.  While it's field marks don't readily separate it from other dark-rumped petrels, it's unique structure and flight style combine to give it a distinctive character or 'jizz'.  But it's not at all easy to convey that concept in words, let alone to capture it in a still image.  Certain video grabs from this video will contain just the right combination of field marks and structural cues to allow an accurate identification.  Others will convey very little or may even throw the identification completely off target.  What is it that we need from a digital image to be able to get everything just right?



Digital images freeze gestalt as it were.  But, for various reasons we often struggle to make sense of what we are actually seeing.  The original video footage of this Bulwer's was captured using a handheld camcorder in full optical zoom on board a fast-moving sailboat.  I devised a technique to deal with the horrendous camera shake in the footage (HERE) and this animated gif is the result of that effort.  The process involved locking each image around a single locus.  You will notice the bird's eye doesn't change position on screen - that was the key to presenting this bird's features and it's gestalt to full effect.  Could there be a similar key to allow us 'lock down' not just a birds position on screen but it's structural traits - i.e. the essence of it's gestalt, in a single digital still image?  The key it transpires is to find a locus or loci that help define a bird's distinctive morphology and lock them down using a special tool which I call a Gestalt Key.


When we delete all field marks and strip back a bird to its basic shape we start to get a feel for an identification based upon gestalt.  But we also give too much freedom to the brain to latch on to anything that it thinks we can use to make a judgement call - and as a result we frequently get our analysis wrong.  Have you guessed what these four bird's in this image might be?  Answers can be found HERE.

We may use comparative morphology techniques and even 3D modelling (explored HERE) to conjure up and test ID theories to some extent.  But does any of that ever actually get us any bit closer to a firm identification?  Or, do these tools simply reveal how little our digital images convey about a bird's true shape and gestalt.  How do we actually nail down an identification more firmly?

Throughout this blog I have explored many of the classic tools birders employ to try and measure morphological features from digital images.  Without exception I have found all the current techniques to be limited in their accuracy.  There are a number of reasons for this.  In the posting on primary projection (PP) for instance (HERE) I observed that the challenge to measure PP correctly stems from among other things, the anatomy of the avian wing and subtle movements of the 'arm' and 'hand' relative to the body.

Similarly, while reviewing leg proportion analysis (HERE) I found a lot of the difficulty with that technique once again stems from the bird's hidden morphology.  Much of a bird's leg anatomy actually resides inside the body cavity.
But perhaps the most important challenge, more important than any anatomical movement is the problem of perspective and foreshortening.  All digital images are merely two-dimensional projections of a complex three-dimensional world.  We can only hope to take accurate measurements of lengths and angles from the X and Y axes which constitute our 2D image.  We have no way of measuring anything in the Z axis that projects from the camera's sensor out into the world.  Time and again foreshortening limits the accuracy of our conventional measurement tools.  It is very important to realise that it is not simply a matter of obtaining an image of a bird in profile.  The real problem is that none of the features we try to measure are ever truly in profile at all.  Afterall, these are all three-dimensional structures.  Take for instance PP.  When the primaries are in side profile with the camera, the wing is normally bowed slightly, rarely flat.  More importantly the tertials and tail which may be used as part of the PP calculation are both skewed at different angles relative to the primaries.  So nothing is measurable on a perfectly flat plane.

Similarly when we try to measure tibia/tarsus ratios chances are these are both splayed at subtly different angles from the body, rarely perfectly in profile with the camera or in the same plane as one another.  Or if for example we are trying to compare the ratio of a bird's bill length or width relative to it's eye proportions, once again these structures are never both in profile together because a bird's head is conical in shape, not square, so it's eye is always offset at an angle from the bill.

The challenge is the same, whether we are trying to measure the lengths of structures or angles created by them.  Perspective alters everything.


So, to summarise, we face the problem of trying to lock down highly dynamic, three-dimensional loci in the very restrictive two-dimensional plane that we call our image.  After the realisation that it is simply impossible to control all these loci, comes the reality that what we need is a different approach to looking at all these problems.

Gestalt Keys
The solution I have come up with is the Gestalt Key (GK).  The concept is simple enough. Firstly we must design a partially transparent mask or stencil which we incorporate as a layer over our image.  We can do this using any of the image editing software programmes (Paint.net is freely downloadable).  The mask contains key points which I refer to as loci.  These are key to identifying a bird using gestalt.  This mask or stencil is our Gestalt Key.  For the purpose of this posting I have designed two gestalt keys following on from my most recent posting on structural angles.
So, lets give it a go.  In the video below I present one of these two prototype keys - the Dowitcher Key for measuring Loral Angle.  Over the course of the next few postings I hope to bring a number of keys up to a reasonable standard and make them freely available for free download and testing.


Here I am using gestalt keys together with structural angles.  I intend creating keys that can accurately and consistently measure primary projection, bill to eye ratio, leg proportions etc.

Conclusions
This posting is the culmination of much research, trial and error.  Techniques for assessing the morphology of birds based on digital images are not a new concept, but what has been lacking invariably is a rigorous tool to help narrow and possibly even eliminate margins for error in measurement.  I hope Gestalt Keys will help bridge a gap and make this form of analysis just as valuable as the analysis of fine feather detail from digital images.  I hope this is the key to finally cracking gestalt.

Wednesday 6 April 2016

Gestalt - The Limitations of Structural Angles

As all the postings in this series have shown, gestalt or 'jizz' (also spelt J.I.S.S.) is a particularly difficult thing to measure.  In many ways, the strong proponents of a 'jizz-based' approach to identification have been dealt the short straw as it were.  Gestalt is exceptionally difficult to define, to describe and to nail down in terms of scientific measurement.  

Many of the postings under the subject banner Gestalt have contained the words "the limitations of...".  I honestly set out studying this branch of the blog with the belief that I could tie down gestalt using some of the commonly applied tools such as primary projection, bill to eye ratio, leg proportion analysis, beak structure and shape, comparative morphology, etc.  So far it seems that any real attempt to define the morphology of a bird based on digital images is fraught with difficulties.

There is one popular identification tool that I haven't really given much consideration to before now.  Occasionally one comes across the use of angles evident in bird morphology for identification purposes.  I am  not sure what the generic term is for these angles so I refer to them collectively here as Structural Angles.

Gular Pouch Angle (GPA)
Many experienced birders would probably have heard of GPA or 'Gular Pouch Angle' used in the sub-specific identification of Great Cormorants (Phalacrocorax carbo)  here in Europe including the Atlantic (P. c. carbo) and Continental Cormorant (P. c. sinensis).  


Following my recent successes with 3D modelling for assessing the impact of perspective and angle of view on our subjects, I decided to use this technique once again to test the usefulness of structural angles in bird images.  For this I needed to generate a rudimentary 3D Cormorant.


Based on preliminary assessment the GPA is a useful and seemingly fairly robust feature to assess from photographs.  That doesn't guarantee of course that a bird is assignable to one race or another.  The value of GPA as an ID feature remains under constant scrutiny thanks to the thorny subject of racial hybrids.  For the purpose of this blog however it is interesting to finally find a gestalt-based tool that might actually work quite well.

Can the same be said for other structural angles?  As it happens I am not aware of many other structural angles in regular use and I'd like some feedback if anyone knows of others that might be worthy of analysis.

Loral Angle
Another structural angle that recently came to my attention is the Loral Angle used to assist in separating Dowitcher (Limnodromus) species.  Once again I have taken to generating 3D models to put this one to the test.



Note that in the analysis of Cin-Ty Lee and Andrew Birch (HERE) they observed that the eye in Short-billed Dowitcher is positioned in the head just slightly above the position where it is found typically in Long-billed Dowitcher.  Based on my analysis I am inclined to conclude that the eye in Short-billed Dowitcher should actually be a little more forward in the head in addition to being a little higher, but that's only of minor consequence.  This looks like it might be another really effective tool, but how well does it stand up to some 3D testing?  Specifically how does camera perspective alter the loral angle, with the potential to render a false identification.  In the video below I present some initial findings.



The results of the 3D analysis indicates that the loral angle is heavily influenced by camera perspective.  As the eye in Short-billed Dowitcher appears to be positioned marginally closer to the bill one might expect a discrepancy to occur more frequently while measuring the loral angle of that species.  Worth mentioning of course that, as the authors point out, this feature should only be used in combination with others to form a firm identification.

Conclusions
Structural Angles are an interesting approach to the complex problem of judging gestalt or morphology from photographs.  Foreshortening occurs when three-dimensional objects are projected onto a two-dimensional plane such as a photograph or screen.  This means that only objects which are flat and perfectly parallel with the camera can be measured accurately.  This principal applies equally to the measurement of angles from photographs.  As reliable identification features go GPA appears a bit more robust than Loral Angle but both tools clearly have their limitations.  I believe there may be a slightly more accurate way to approach Structural Angles from photographs.  I will elaborate more in a forthcoming posting.