Thursday, June 5, 2014

Not all images that are 3D-enhancible have 'encoded 3D information'.

What was done to the Shroud image can be done to our flags too, with a few clicks in image-enhancement software programs (in this instance ImageJ).

The Turin Shroud image is famously 3D-enhancible, given the right software. Initially it was shown with the so-called VP8 image analyser that was allegedly space-age technology, and not surprisingly led to much over-hyped speculation that the TS image was different from any other. (Caveat: the article has "multiple issues" according to wiki. I would concur with that, it being less about the VP-8, and how it works, and arguably more a vehicle for "Shroudie" propaganda).

This  blogger pricked that particular balloon some 2 years ago, pointing out that the 3D- enhanced images not only brought the man's image up out of the page, but the 1532 scorch marks as well. (Wikipedia credits me with making that finding, but I'll try not to let it go to my head).

There's more talk right now about what the modern day digital equivalent to the analogue VP-8  (notably the freely downloadable ImageJ software) does or does not do to the TS image that is meaningful. In other words, what are the 'correct' settings that gives a valid image?

Again, I've expounded on that question in the past, and shown how one can 'normalize' ImageJ settings by taking a scorch imprint from a bas-relief metal template, and then enhancing it in ImageJ to get the closest possible match to the original. If one then applies those same settings to the TS image, one gets a very realistic-looking semi-3D image of a bearded man.

("Normalised" 3D-enhanced image from a 2012 posting - see link below)

But inevitably there will be those who reject that approach, denouncing it as self-serving, a means of selling my "scorch" hypothesis.

Are there other ways of deciding on the optimal settings?

Here's a simple experiment one can do that brings home the arbitrary nature of ANY settings one applies.

Select any flat, non-3D image that shows a gradient of colour intensity. The one selected here is a colour palette, aka shade chart,  from MS Paint (click on the menu's Custom colour).

Now enter that image into Image J.

Here's one result that shows a gentle 3D enhancement that reflects image intensity:

So there is no question that the 3D enhancement is entirely artefactual, inasmuch as there was never a 3D object to begin with, simply a planar colour chart.

What if the image intensity rises to a peak, and then falls back again?

Above you see the (fairly predictable) result of overlapping two shade charts, end-to-end, then 3D-enhancing in ImageJ.

Incidentally, the ImageJ settings used were approximately those that I used many moons ago to get the best match between a horse brass (a handy bas relief template) and its scorch imprint on fabric.

Left: original brass. Centre: scorch imprint (left-right reversed). Right (with artificial colour which seemed a good idea at the time): optimized 3D enhancement in ImageJ intended to achieve the closest match with original, thus establishing 'normalized' settings.

Yes, they were what I called 'normalized' settings and which later produced that pleasing result shown earlier with the Shroud Scope/Durante 2002 Man on the TS.

Link to my March 2012 posting

But those same settings can be used to endow many 2D images with 3D or, more correctly,  pseudo- or virtual 3D character.

A week ago, working at home on my main computer, I constructed the chart as a series of concentric circles, with lightest shades on the outside, becoming darker towards the centre. When entered into ImageJ the diagrams were converted to pleasing cones that telescoped so to speak out of the page. The model system allowed one to see precisely what effect could be achieved by twiddling the various controls, especially z (for more or less 3D) and smoothing (which in simple terms can be thought of as the 3D equivalent of converting a bar chart into smooth curve). The fun part was to create irregularities in the sequence of shade intensity, so instead of a smooth progression like 1,2,3,4,5,6,7,8, one had 1,2,3,6,4,5,8,7.  One might have expected the relief to respond up or down with each change in shade. but that was not the case. Abrupt changes of shade in the sequence 'forced' the overall relief such that expected micro-changes elsewhere in the local topography were suppressed.  Sadly I did not save to memory/thumb stick, but will add the graphics onto this posting (and more besides) later when available.

Update: I've been able to produce the same effect on this older travel laptop in MS Paint, despite its Windows Vista having less functionality in Paint than Windows 7.

Here's a series of images that hopefully speak for themselves obtained just now when a colour-graduated series of closed contour lines are manipulated in ImageJ. (Why is it I can't get Madonna out of my head right now?)

2D image constructed in MS Paint

After applying ImageJ with low z axis value

Now with high z value

Low smoothing

Increased "lighting" (lateral, thus creating artificial shade effects)


Take away message for now: 3D enhancement of a 2D image does NOT necessarily mean that the image has "encoded" 3D information. It may simply have gradients or even abrupt steps in colour intensity which the program reads as more or less height on an artifically imposed z axis.

Joseph Accetta PhD, electro-optics/laser/imaging expert said as much at the 2008 Ohio Shroud conference, though you have to go patiently page by page through  his now hard-to-find Power Point presentation. See yesterday's Dan Porter posting, and an acknowledgement to this blogger for having unearthed it.

Further update:  Well, I've persevered, despite MS Paint not allowing me to create nests of concentric circles in Windows Vista, so I've had to make do with ovals. From earlier you may recall that the aim was to show the kind of anomalies that can arise when the stepped density gradient values are out of sequence.

Here's the 2D starter, with an irregular sequence of image density.

Here's the first result in ImageJ, where one sees that the program has failed to respond to the density gradient at the peak, due to there being a dominant intermediate zone of image intensity. But tweaking of settings, especially of z value and smoothing, can make the program respond to the peak, as the next graphic shows.

The peak within the 'crater' is now visible, but the overall image has a poor 3D enhancement, due to that out-of-sequence gradient of image intensity.

Late postscript: followers of this and my specialist Shroud blog will know who I believe is depicted on the Shroud of Turin via some kind of thermal imprint (maybe chemically-assisted). It's a historical figure, alright, but not Jesus Christ. Who?  Probably Jacques de Molay, last Templar Grand Master, executed on the banks of the Seine in central Paris by slow-roasting in 1314.Others before me (notably Knight and Lomas) have arrived at the same conclusion, even if their proposed imprinting scenario is totally different, involving in their case a still live subject, as distinct from bronze effigy or similar.

Here's an image of de Molay I've just found by googling, fashioned and described as a 'mascot' (don't ask me why).

Let's see how that 2D image responds in ImageJ. Does it  have what some might misleadingly describe as "encoded 3D information"? I leave you to judge dear reader:

Some of us get a feeling of déjà vu  when viewing artistic portrayals of Jacques de Molay, even modern-day ones as here.

Update: Sunday 8 June

In response to points raised elsewhere I may make additions or clarifications here, but won't be drawn into debate that has a backbiting nature, based on this free spirit being seen as one of "them" rather than one of "us". I'm only here for the science.

The question has been raised as to whether colour can endow 2D graphics with superior 3D-enhancibility relative to the TS image (the latter invariably viewed as the gold standard needless to say, whose alleged uniqueness is assumed, not posited).

I've just this minute taken one of my earlier graphics, the shade chart from MS Paint, and converted to gray scale before uploading to ImageJ. Here's the result - identical except for colour.

Now before exploring the substitution of gray scale for colour, here's a comparison which I bet you've never seen before of the behaviour of (a) the TS face in ImageJ (or similar) alongside (b) a model scorch (c) a model shade chart and (d) a painted portrait of our old friend Jacques de Molay. I'm showing it first WITHOUT and then WITH a tick in the ImageJ "Invert" box. Absence of a  tick brings up the denser image regions (seemingly) above the plane of the screen, whereas a tick in the box puts dense regions in hollows below the plane. Note especially the response of the shade chart, third from left, producing a cone without the tick, and a crater with. Click or double-click on images to enlarge.

There are some interesting differences there that need discussion, but let's deal first with the matter of gray scale, and see if that makes any difference to what you see above.

Please be patient. More to follow later today.

Resuming 12:49 pm local time:

OK, here are the gray scale versions of the two graphics immediately preceding this update.

Conclusions: both the TS image and the model scorch were enhanced better without a tick in the Invert box, such that the shade chart appeared as a peak rather than crater. But it seemed that the portrait responded better when there WAS a tick in that box. In other words, imprints, whether of 'mysterious' origin (the TS) or scorch imprints from a hot metal template seem to be qualitatively different from painted portraits in their response to image enhancement in ImageJ. (underlined words  unfortunately omitted from first draft)

All the tests so far were done on the 'as-is' images, without any prior "photographic inversion", i.e. reversal of light and dark, as per Secondo Pia. Does prior photographic inversion alter subsequent response in ImageJ, tested with an without a tick in that (somewhat confusingly named) "Invert" box. where invert in that instance means raising or lowering the vertical z axis above or below the viewing plane (something different from photographic inversion from "positive" to "negative" and vice versa.

More to come later in the day.

So, no surprises  there then. The monochrome results mirror the coloured ones, and graphics that are black/white reveresed give the same result, provided the Invert box is ticked or unticked to allow for the switch in tonal distribution.

The chief finding in these latest experiments is to confirm what I found over 2 years ago, namely that the settings in ImageJ that work best for 3D-enhancement of the TS, judged artistically, are essentially the same as those judged more objectively by matching up the 3D-enhanced scorch image with the template from which it was derived. You would not know this from reading much of the Shroud literature where, time and again, after searching diligently for the few references dealing with  model scorches, one finds derogatory references to their quality and/or response to 3D imaging, compared with the Shroud image. One also finds claims that the Shroud image is 'uniquely' responsive to 3D-imaging, which one sees above are patently untrue. There seems little to distinguish between the TS and a scorched-on image: STURP almost said as much in its 1980 summary.

Postscript: I have just this minute added a section to my earlier posting on the TS man's toes, those very elusive toes, deploying a new (desperate) ruse to improve 3D-imaging in ImageJ. It's been, well, moderately successful. 

Final gasp (on this over-long posting):  there was a curious and unexpected feature of those 3D images where 4 different images were tested together. Switching between Invert On versus Off in Image J did not produce so marked a transformation as expected (one expects the image to be turned inside out, like punching a hat to make a new one with the lining on the outside).

Here's what happens if one performs the switch on the TS image on its own:

Well, I never. The result is surely different from testing them singly. Must go and check the sequence again, make sure there's no mistake..

Nope, no mistake. On a quick glance they looked similar, especially if one holds one's laptop at arms' length. But they are clearly not.  Note the guide to "sense" in a topographical way - the convex pucker at the top of the head on the left, the indentation at the same location on the right. The ease with which the two can be confused is probably due to the relatively small z value used to get a pleasing result, giving relatively shallow relief above and below the base plane.

OK, let's compare the TS and scorch side by side with HIGH z setting to exaggerate relief, and then toggle between INVERT on and INVERT off.



Note the identical response of the two images in Image J under these different settings. Go figure, as they say.


Hugh Farey said...

Terrific. Love it.

sciencebod said...

Thanks Hugh. Do you like the flags too (just added a second ago as an afterthought)?