History of the UnBlur algorithm

In this page I provide a brief account of the history of my development of the UnBlur algorithm.

November 1999: A summer break diversion

In November 1999 I had just finished my second year of full-time work as a high school Mathematics teacher at Mentone Grammar, a private high school in Melbourne, Australia. (The academic year is the calendar year in the Southern Hemisphere; it begins after the summer break, as it does in the Northern Hemisphere.) I had secured that job following half a year of part-time work there during the second half of 1997, the year that I had returned to the University of Melbourne to obtain a Graduate Diploma of Education, the qualification required for any unversity graduate to teach at any school in Australia. (I had just finished a postdoc in theoretical particle physics in the Department of Physics at the University of Melbourne, but hadn’t made the right connections to secure a subsequent postdoc position; high school teaching was going to be my “making ends meet” job until I secured another physics position.)

I was looking forward to a bit of a summer break. My (at that time only) son was a year and a half old, and we were finally back on a secure enough financial footing that I had just been able to quit my second job as a part-time video store clerk (20 hours a week at $10 an hour). So I went down to our local library, in Narre Warren (a suburb of Melbourne), and browsed the shelves. There I spotted a coffee table picture book, The Killing of a President, by Robert J. Groden. I had seen that book six years earlier: a young brother of a then-girlfriend had shown it to me. (Yes, I wondered what a 12-year-old was doing with a book with gory, graphic images of JFK’s head, but anyway.)

I decided that it would be a perfect diversion for relaxing over the summer school holidays (well, it is the prototypical definition of “True Crime,” right?), and borrowed it.

As I was looking through it, I noticed how many of the historical photographs shown were blurred (generally through motion of the camera, or of the presidential limousine, or both). I remembered reading that frames of the Zapruder film had been “de-blurred” by physicists at Los Alamos back in the ’70s (although I was somewhat confused by the fact that many of those shown in Groden’s book were still blurred). But these other photographs may never have been attempted. I wondered: how difficult would it be to deblur them?

I knew that the standard way of attacking such a problem would be to convert everything to Fourier space, which converts the blurring convolution into simply a multiplication of Fourier-space representations. Deblurring “only” requires dividing back out the Fourier transform of the blurring—i.e., that of the “point spread function.” The catch, of course, is that this Fourier-space representation has zeros, and you can’t divide by zero. The general way to work around this was to model the noise in the image, and then divide not by zero, but by something nonzero that took into account the noise.

I was wondering if there was an alternative way of attacking the problem: directly in the original photographic space, rather than in Fourier space. Such an algorithm would then need to deal explicitly with convolutions, so it would not be nearly as computationally efficient as a Fourier space algorithm. But I thought it would be fun to try. (I was also still performing some research in theoretical physics: my final publication in physics, in the American Journal of Physics explaining a subtle aspect of Einstein’s theory of relativity—including possibly the only appearance of the starship U.S.S. Enterprise in that esteemed journal—was submitted in February 2001.)

I scanned some of the photographs in Groden’s book with my color scanner, and on January 7, 2000, I started writing some C code to try out ideas.

2000: Different ideas, and the Zapruder film

I played with what I was calling “UnBlur” on and off over those summer holidays. By the time I returned to start the 2000 school year in February, I had shelved those ideas for the time being.

By August I was again tackling the problem. I decided I would ask on some Usenet groups on the internet (the way that folks discussed things in the early 1990s, before there was a World Wide Web or much on it to help; I still thought the Usenet to be a good source of information in 2000) whether all of the frames of the Zapruder film were actually available anywhere. I was told that it was copyright, but that there was legislation in the early 1990s that had mandated the creation of a reference copy for public use, and it had been released on DVD by a company called MPI in 1998. I didn’t have that DVD, but an “assassination researcher” (a creature I learned existed), David Wimp, generously emailed me the frames, captured from that DVD, one by one.

I started testing the UnBlur program on some of those Zapruder frames, as well as some of the other photographs I had scanned from Groden’s book.

By October I had overhauled how I was approaching the UnBlur program. I then returned to it while attending the 2000 Congress of the Australian Institute of Physics in December. (Mentone Grammar was kind enough to support my continued work in physics, as an unpaid Research Fellow at the University of Melbourne, by supporting my attending the biennial conference of its professional body—even when it fell in the summer school holidays, as it did in 2000.) I overhauled my approach yet again in January 2001, while still on school holidays, focusing on how to back-propagate the error in each successive approximation.

2001: Something workable, and more JFK

I again returned to my UnBlur program in June 2001, after a hiatus of some months. I improved the back-propagation further, and introduced a relaxation (or annealing) step. By late August I introduced a method to accelerate convergence.

In the mean time, I had been made aware of some private forums on the internet dedicated to the JFK assassination. I learned that there were two main forums: on one, run by Rich DellaRosa, people actually believed the Zapruder film to have been altered, or completely fake; on the other, JFK Lancer, folks generally believed the Zapruder film to be completely authentic. I stayed out of that debate, at the time, but provided deblurred frames, and even video clips of sequences of deblurred frames, for discussion.

Finally, I decided that my UnBlur program was probably as good as I was going to get it, at least for the time being, and on Sunday, September 9, 2001, I released a first version of it on my website.

A well-meaning but silly offer

Two days later, while correcting Mathematics tests late into the night, the television that my stepsons had left running in the background on Channel 10 (after watching The Simpsons earlier in the evening) burst into Breaking News with Sandra Sully about a small plane crashing into the World Trade Center in New York. Being 10:50 pm at night (8:50 am in New York), and with a six-month-old second son not giving us much sleep through the night, everyone else was asleep. I woke up my then-wife, and we went back out to watch the coverage. “Another plane just flew in and crashed into the building!” I yelled. Even the commentators—who were talking by telephone to someone in a nearby building—didn’t realize what was going on: the person on the phone had yelled something like, “There’s another explosion; I’m getting out of here, sorry!”

Thirty-six sleepless hours later (I watched the events transpire right through the night, on two televisions on different cable boxes next to each other—which we fortunately had installed just feet apart from each other—and I even carted a portable black-and-white television into my classroom in the morning, so that both my students and I could keep track of what we thought might have been the start of World War III), I sent an email to the FBI in the United States, offering the use of my new UnBlur program, should it be of some use to them analyzing the frames of video footage of the event. In retrospect, it was a silly offer; but it somehow made sense to me at the time.

Needless to say, I did not receive a reply.

2001–2003: The Zapruder film

My introduction to the controversy over the authenticity of the Zapruder film led to my going down that rabbit hole in 2002, and Professor Jim Fetzer invited me to present my findings—whether they supported authenticity or not—at a conference at the University of Minnesota, Duluth, that he was planning for May 2003. I accepted, and he most generously had shipped to me two boxes of books about the assassination, and particularly the photographic evidence.

My consequent work on the Zapruder film is well-documented elsewhere on my website, and you can view my 2003 presentation via the YouTube links there.

Ironically, my work on UnBlur, that got me onto that case in the first place, played an almost insiginficant role in my work on the Zapruder film. The one exception was one of the deblurrings that I did in 2001, while still developing UnBlur, in which I alternated an UnBlurred Frame 313 with a precisely overlaid Frame 312 (a better 2021 version is here), which demonstrates most clearly that the Zapruder film depicts the president’s head moving forwards between Frames 312 and 313. This actually attracted the attention of Josiah Thompson, the author of the seminal Six Seconds in Dallas in 1966 which analyzed this forwards movement in great detail (and who in 2001 was adamantly opposed to Jim Fetzer’s claims about the film being inauthentic). Thompson had latched onto an erroneous analysis by David Wimp (the person who had, ironically, originally emailed me all of the frames of the Zapruder film!) that had originally concluded that the blur in the film meant that it did not actually depict the president’s head moving forwards. Wimp issued a correction shortly afterwards (i.e., back in around 2002), but Thompson repeatedly ignored the retraction.

Remarkably, it took another 20 years for this to come to a head: Thompson’s new book, Last Second in Dallas, finally published in 2021, and relying critically on Wimp’s original erroneous conclusion, has been completely discredited by Wimp’s own insistence that he had retracted that conclusion about the head movement nearly two decades earlier, and could not understand why Thompson built an entire theory of the film around it. It was somewhat satisfying that my original UnBlur analysis of Frame 313 was, two decades later, able to provide such a graphic and easy-to-understand demonstration of the incorrectness of Thompson’s theory—and that my original generous contributor of Zapruder frames, David Wimp, would confirm that Thompson was actually confused and wrong the entire time.