There’s always room for improvement, as they say. In February 2014, I posted my first entry on this blog, “Historical Stereoviews as Tweened Animated GIFs,” demonstrating a method of creating animated GIFs from stereoviews that aren’t as headache-inducing as the common “wiggle GIF.” You might recognize our alligator friend from that post—it was my first-ever experiment along these lines. The technique I introduced then consists of lining up two stereographic images in Photoshop and then using the “tween” function to fade gradually from one to the other and back. Wherever the two images line up, as with the alligator’s midsection, this yields a pretty compelling illusion of depth and motion. However, wherever the images don’t line up, as with the alligator’s head and tail, there’s a discontinuous jump from one position to the other. I knew it was possible to interpolate frames in such a way that every part of every object would transition smoothly from position to position, since I’d seen it done in a YouTube video, but that work had been carried out using specially-designed software that seemed to be beyond reach, since the website associated with it (Start3D.com) had been taken offline. Fortunately, I’ve now found another way to achieve transitions of this sort using readily available freeware. Here’s our alligator friend again:
A little better, isn’t it? I created this second animation using FotoMorph, a free image morphing program. Judging from how it’s promoted, FotoMorph seems to have been designed mainly for morphing pairs of dissimilar images, such as one person’s face into another, but it also lends itself nicely to stereoscopic pairs, as my alligator shows. The process is more labor-intensive than simply lining up two images as I was doing before, since it involves associating multiple corresponding points in the two images by hand—the procedure is to click anywhere in one image to create a dot there as a new reference point, and then to click and drag the dot that appears in the other image to the corresponding point. Here are the hundred or so points I had to define to create the alligator animation:
Moreover, it’s not just a matter of associating points that correspond to each other; consider cases where foreground objects shift position relative to background objects, or where part of an object visible in one image isn’t visible in the other. Dealing strategically with situations like these can get very complicated very quickly, and fixing one part of an animation can unexpectedly cause glitches in another, seemingly unrelated part of the animation. I’ve been able to resolve many cases of “incorrect” morphing, but sometimes I’ve just given up in frustration—and now that I know what to look for, I see similar problems throughout the Start3D YouTube video, so I guess I’m not the only one who’s struggled with them.
On the bright side, morphing can yield pretty gratifying results when it works correctly, whether due to effort or luck. Take this stereoview where one image is zoomed in relative to the other (compare the positions of the feet):
If we were simply to overlay the two images as in my earlier approach, we wouldn’t be able to line up more than one small piece of the scene at once—a head, say, or maybe the hands. Fading from one image to the other would just cause the other parts of the scene to blur. By contrast, morphing between the two images produces a nice illusion of three-dimensionality throughout the whole scene:
We can also use morphing software to animate images taken from vantage points that weren’t side by side and so don’t lend themselves to stereoscopic viewing. Here’s a pair of tintypes taken through two lenses mounted one above the other, like the third pair of examples in my earlier blog post on “‘Twin’ Tintypes as 3D Animated GIFs”:
Of course, we’re not limited to working with simultaneously-photographed images such as the ones I’ve presented so far. Photographs taken at different times can yield equally striking impressions of three-dimensionality and motion through image morphing as long as they’re sufficiently like one another for us to be able to interpolate intermediary frames, even in cases where the gaps would otherwise be too wide to produce an acceptable illusion. As a case in point, here’s a pair of photographs by Félix Nadar of cartoonist Amédée de Noé (“Cham”), circa 1870, which I scanned from a facsimile in Nigel Gosling, Nadar (London: Secker & Warburg, 1976), page 217:
Nor are we limited to morphing between pairs of images; we can extend the same principle across as many images as we want. Here’s a “Portrait of a Woman in Nine Oval Views” by Southworth and Hawes, the acclaimed American daguerreotype artists, which I grabbed from Wikimedia Commons:
It’s not apparent what order these nine “oval views” were originally taken in, but I’ve numbered them above in a sequence that minimizes the change in facial angle and perspective from step to step. If we rapidly display the nine views in that order, we get this:
There’s some illusion of three-dimensionality here, but once again it resembles the “wiggle GIF” in its jerkiness. It also rolls by too quickly for the eye to get a decent visual fix on anything, but the more we slow it down, the more conspicuously jerky it becomes. Fading from view to view would conceal the abruptness of the transitions, but only by blurring them. By contrast, image morphing can give us a smoothly continuous animation at any speed we choose:
Granted, this isn’t perfect. I couldn’t get the nose and cheek to morph properly across the gap separating view #7 from view #8, so I left out views #8 and #9. There are obviously some issues with the neck and shoulders as well, since they don’t follow the same rotational trajectory as the head. Overall, though, I think this is a solid improvement—and a pretty cool animation besides, given that we’re looking at the face of a woman circa 1850.
If image morphing software can reduce jerkiness in cases like these, it stands to reason that it ought to be able to do something similar for “real” motion pictures with low frame rates. Here’s a straightforward animation based on sequence D of Plate 488 from the second volume of Eadweard Muybridge’s Animal Locomotion, published in 1887, with the stairs held in as constant a position as I was able to manage:
Muybridge ordinarily states the time interval between frames in his separate Prospectus and Catalogue of Plates, but Plate 488 is an exception, with its listing keyed to the note: “No record of intervals of time between phases.” The sequences on it also lack the usual background reference grid, so maybe this was one of Muybridge’s earlier plates, produced before he had fully worked out his methods. In any case, the duration of each frame is set to 0.15 seconds in the above animation—that might be a little slow, but I don’t think it’s implausibly slow. Displayed at that speed, the movement appears very jerky, just as we’d expect from a frame rate of 6 2/3 frames per second. To try to improve the illusion, I used FotoMorph to increase the number of frames from twelve to forty-five (interpolating three frames between each existing frame) and then did a little manual retouching to mask distortions like this, which I couldn’t figure out how to eliminate from the morphing itself:
This approach admittedly assumes a constant, regular rate of motion between each of the original frames, which I’m sure doesn’t quite match reality—although at this speed I don’t know whether the eye can tell the difference. Under the right circumstances, we can extend the same principle to interpolate much longer durations. Here, for example, are three daguerreotypes taken by Southworth and Hawes of the Bigelow School in Boston circa 1850, as reproduced in Young America: The Daguerreotypes of Southworth and Hawes, page 465:
The changing positions of the hands on the clock show that these daguerreotypes were respectively taken at 3:45, 3:49, and 3:50 (remember that daguerreotypes are mirror images). So I created an animation that morphs from the first image to the second over the course of four minutes, and then from the second to the third over the course of another minute:
This video may not be very interesting to watch, but it shows that it’s possible to use image morphing software to recreate “motion pictures” of scenes based on multiple photographs taken at known intervals. The hands of the clock actually “move” at about the right speed, albeit so slowly that it’s not apparent if you watch the video in real time (and even sped up, it isn’t as clearly visible as it might have been if I’d had higher-resolution copies of the three daguerreotypes to work with). Overall, what we’re seeing here is presumably comparable to what a surveillance camera would have recorded if it had been capturing continuous video of the same scene during the same five minutes. With the right source material, I could imagine recreating the motion of clouds or shadows in much the same way.
So what else can we do by applying image morphing software to historical photographs?