Trail Cam Captures Redux: Beyond (My) Reasonable DoubtPosted: August 31, 2018
I’m making another departure from my posting schedule for reasons that I hope will be self-evident. This is the strongest statement of my views I’ve made to date; I’m calling it as I see it based on this new way of looking at the photographs. As always, click on the images for better viewing.
A Louisiana photographer (and new Project Coyote member) who has been following our efforts, reached out to me in response to the recent trail cam capture and alerted me to a feature in Photoshop. This feature makes it possible to stack two images and subtract the identical pixels, leaving the differences between the two images visible in the new composite.
I understand that this is sometimes referred to as subtracting and sometimes as ghosting. Given my limited knowledge and skill set when it comes to image processing, this was a revelation. The method is described between minutes 1 and 2 of this YouTube video. (The rest of the segment pertains to image stacking for landscape photography and is not relevant to this discussion.)
He sent me three examples – the recent capture, the neckbird, and a pair of photographs showing what Frank Wiley had identified as a female Ivory-billed Woodpecker in flight. I didn’t get any new insights from the last of these; the results for the most recent capture were interesting and possibly significant; the neckbird was a revelation, eliminating any lingering misgivings I had about the image and convincing me (at least) beyond a reasonable doubt that it shows an Ivory-billed Woodpecker.
Bear in mind that the images discussed below are among a handful of captures (out of well over a million images obtained since 2009) that we’ve found to be suggestive or better. (At the 20 second interval used on our current cameras, a single month’s 12 hour-a-day deployment will produce over 60,000 images; the Reconyx cams had a 30 second interval, but that’s still over 40,000 per camera per month.) So it’s not as if we’re finding possible ivorybills everywhere, let alone frequently.
I recently started using a different, ostensibly more user-friendly, image-processing program, Luminar, and I found that it has the same capacity. I tried it on the other oft-discussed trail cam capture with the same revelatory result, although in that case, I did not have access to both unprocessed originals and had to use one cropped image and another that had been cropped and lightly processed for brightness and contrast. I don’t think this impacted the results. I used Luminar to process all the images discussed below.
Immediately below is the original “Neckbird” capture and an enhanced crop, followed by the processed image, showing the difference between this frame and the one before it. Next is a detail, showing the ‘ghost’ outlines of the image that appeared in the original. (Note that a second bird may be present in the form of a bill protruding from the lower cavity in the snag at right, as discussed in this post.)
For me, the most startling result is that the bird looks even more serpentine and elongated in this iteration. This is mostly because the lower third of the body (apparently including part of the tail) is better resolved and distinguishable from surrounding objects. What I had thought was a patch of white near the base of the bird, just above the lower diagonal branch, turns out to have been a leaf. I think this longer, leaner profile is inconsistent with Pileated Woodpecker and consistent with Ivory-billed Woodpecker, and not just because of the neck, the feature that was most suggestive to many who viewed the original capture. I’ve seen a lot of Pileateds and have looked at countless images online, and nothing about this better-resolved silhouette suggests Pileated. The remnant clips from the 1935 Ivory-billed Woodpecker film may be illuminating in terms of neck and body shape.
The silhouette in the November 30, 2009 capture, which was the first suggestive image we made public, is not as compelling as the one in the image shown above, but the capture shows three apparent ivorybill field marks – a black crest, white on the lower part of the bird, and a large, light-colored bill. In this case, the processing provides additional information about these apparent field marks. The bird also appears to be larger than a Pileated Woodpecker. The basics of all these issues are addressed in Frank’s original discussion of this capture.
To reiterate, I am not in possession of the original frame that follows the capture. It was probably lost after Frank’s passing, so I used two crops for the first composite, one showing the suspected ivorybill and the other, the subsequent frame. The crop of the empty frame was otherwise unaltered, but the one with the bird had the brightness and contrast adjusted. The output differed slightly between the “difference” version and the “subtraction” version, something I did not observe when working with unprocessed images. The difference, however, appears to be limited to background features and not the bird.
Remember that this capture was obtained from a spot where I’d had a sighting of a large woodpecker with a lot of white on the wings on the 25th . Both Frank and I heard the wingbeats, which were loud and distinctive. This was a rare instance in which a bird appeared to have been brought in by a double knock series. We’d also had an auditory encounter about 300 yards away on the morning of the 24th, while staking out the cavities in the willows shown above.
The first image shown below is the original capture with the red box around the bird. The crops appear as a pair of images. The first shows the bird, and the second is the subsequent frame with the bird absent. Those are followed by the difference and subtraction composites. More discussion after the images.
As I read this imagery, I think it suggests that the apparent white saddle is part of the bird, not light leaking through or an intervening object; indeed, the intervening branch is revealed in the ghosted images, with the body of the bird still visible behind it. That said, this may be the most ambiguous aspect. The treatment also highlights the presence of the crest and shows it is part of the bird; there is no red in the original; if any red were present, it should be at least faintly visible at this range and under these conditions. This interpretation is based both on the color in the background and on a Pileated captured perhaps six feet farther from the camera during this deployment (in an overexposed image included in Frank’s original analysis).
Most important of all is the fact that both the difference and subtracted versions confirm that the apparent very large bill (which appears to be light-colored in the original picture) is indeed a bill and not some other object. Even allowing for some motion blur, the bill is disproportionately large and therefore inconsistent with Pileated Woodpecker. Indeed, all discernible features in this photograph are consistent with female Ivory-billed Woodpecker and nothing else. I have always believed this picture showed an ivorybill. Based on these composites, I am now firmly convinced that it does.
I also tried creating a composite combining the capture, which was taken in the late afternoon of November 30, 2009, with another frame taken in late morning on the following day. The results are somewhat less distinct but the same elements are still apparent.
Finally, I tried this approach with the images discussed in the previous post. The results of this effort were somewhat less satisfactory. I assume because the originals are lower resolution and therefore contain fewer pixels. Lighting conditions may also be a factor.
This capture offers a lot less to work with. The Reconyx jpegs are 2048 x 1536 or 3 megapixels, while the Plotwatchers are 1280 x 720 or less than 1 megapixel.
I was unable to find additional specifications for the Reconyx cams, but the folks at Day 6 Outdoors were kind enough to provide them for the Plotwatcher. Per Day 6, the sensor is .25″ (which is smaller than most cell phone cameras); the focal length is 3.5 mm (which I think translates to the equivalent of an ~20 mm lens on a full frame or 35 mm film camera); and the aperture is f/2.6 (apparently the equivalent of ~f/16).
Nevertheless, I experimented with using both the original captures and versions enlarged using Topaz A.I. Gigapixel. I applied the same process to the squirrel capture from a few weeks earlier, which was taken under better lighting conditions. The relevant images and discussion are below, starting with an enhanced version of the original capture with an arrow pointing to the object of interest. Remember that the silhouette that resembles the head and neck of a woodpecker is background vegetation. The object of interest is the small bit of black and white below it.
As mentioned, I’m less confident about interpreting the results for this image, since so much less detail can be gleaned. Unlike the other captures, the subtraction or ghosting does not reveal a clear outline, and aspects of the object which are distinct when moving from frame to frame are obscured in this treatment. Also note the near absence of difference between the preceding and following frames.
In the enhanced iteration, a little more of the outline of the putative IBWO may be visible. I suspect that the reason the white saddle does not appear uniform in this one is due to pixels and that the dark patch below the bright white band is consistent between images because the foreground and background pixels match, not because they’re the same. This explains why the bottom part of the saddle, where the wings meet, is also white. In other words, these are the places where the object was in front of darker pixels. That’s my guess based on shifting back and forth between the relevant frame and the ones before and after. In addition, the white dots above the “saddle” may represent dorsal stripes.
Things change slightly when using imagery from different days. I’m only using the enhanced versions here, but they again strongly suggest that the white saddle is part of the object, not the background. The second frame is the one that follows the squirrel capture discussed below.
The squirrel capture may provide some additional insights into the process’s limitations with lower resolution images. The squirrel, which is clearly visible in the original image, shows up as a narrow strip that’s easily missed if you don’t know it’s there. The second version is a composite using images processed through Gigapixel A.I.
Weather and lighting conditions differed considerably between the two days in question. The camera was pointed in an approximately easterly direction, probably more ENE or NNE; lighting conditions may account for the fact that this image is a little lighter and some small differences are more apparent from this pair of captures. The weather at the nearest station at the time of the squirrel capture was fair with little to no wind; whereas the weather for the putative ivorybill capture, which was taken in the afternoon, was intermittently stormy with winds probably around 10 mph. Under stormy conditions with foliage moving in the wind, there should be more variation between images, not less, so available light seems a likely factor perhaps in combination with number of pixels.
Thus, I’m not sure how much this process adds when it comes to interpreting the most recent capture and how useful it might be with others from Plotwatcher trail cams. (The two older images were taken with a color Reconyx.) But I hope that the cameras are now deployed in such a way that enhancement and differentiating won’t be necessary.
While the approach does not reveal much when it comes to the new image, I think it is dispositive with regard to the old ones. Of course, I make no claims to expertise in forensic image analysis or the technical aspects of image processing, so I am open to being corrected if any aspects of my analysis are flawed.
Like most searchers, I have tended to focus on whether the evidence is sufficient and have seldom thought about what obtaining conclusive documentation might imply. If the two images from the old search area are ivorybills, a number of questions emerge. So let’s assume (or assume arguendo if you are unpersuaded) that these images indeed show Ivory-billed Woodpeckers:
As an initial observation, if I am right about the two captures, I think it becomes overwhelmingly likely that the apparent bill protruding from the cavity in the background of the neckbird image also belongs to an ivorybill, so it seems probable that least two birds were present in the area in 2009-2010. This is consistent with what six of us heard during an auditory encounter in January 2010. One of Frank’s sightings involved three birds. (He was accompanied by the landowners’ teenaged grandsons at the time.) This sighting suggests that breeding might have taken place in the area.
The habitat in the immediate vicinity of the camera trap deployments is not what has been thought of as having much potential; the quality is even lower than I realized at the time. I knew a lot less about forest ecology and conditions in 2009, and while I certainly recognized the habitat as not “optimal” back then, I think I overestimated and perhaps romanticized the overall quality.
Both captures come from within 50 yards of a bean field. The woods in the immediate vicinity did have a number of large trees, and there are good sized cypresses in the area, but the parcel and the neighboring ones (including the one that was clearcut) are not particularly old. Back then someone indicated they’d been cut in the early 1960s.
The site of the earlier trail cam pic and our November auditory encounter is .6 miles from a large parcel of state land that has some very mature forest and inaccessible areas. (The location of my November sighting and the second trail cam capture is approximately 300 yards farther east, away from the public land.) Several other large parcels of public land are within a few miles of the area, so it’s not exactly a suburban backyard. Nevertheless, it’s a far cry from the Singer Tract or our current search area for that matter.
It seems to me that if one accepts that a pair of ivorybills was present in this location in 2009, there’s no basis for doubting the landowner’s claim that birds had been present and using the area for a decade or so. It also supports the idea that audio obtained there included ivorybill kents and double knocks, which tends to validate audio captured elsewhere – Arkansas, Florida, South Carolina, and our current search area.
How does/would the confirmation of ivorybill survival, especially in such ostensibly low quality habitat, impact the assessment of other reports?
Why is that ivorybills are being reported in so many different places over the last two decades? Is it just more publicity, or is something else going on?
Why aren’t there more reports from the public and from experienced birders?
Why are they so hard to document? Even if you read the Project Coyote trail cam captures in the most liberal possible way, there are no more than a handful of images that I think of as likely, out of well over 1,000,000.
Where does one look for them, especially if they are using habitat always thought of as unsuitable?
What is suitable habitat?
What kind of population is out there and how has it sustained itself for so long?
I’m sure there are more, and I don’t have good answers for most. I think the technical limitations of trail cams, scarcity, and wariness (even run-of-the-mill or somewhat overdeveloped wariness) are probably adequate to explain many of the documentation issues but perhaps not the relative paucity of sightings.
I hope to return to my planned posting schedule with the post on range in a week or two and the one on evidence a couple of weeks after that; however, I’ve been looking at composites created from the Reconyx images discussed here. I think the results illuminating with respect to size. This post is already long, so rather than include them here, I will likely do a follow-up examining those images before turning to more general ivorybill-related topics.