Well, I was working on some software to do panoramic image stitching over the weekend. Several hours of work yielded some phase correlation code that seems to work well about 50% of the time, and fail tragically 50% of the time. My brain wasn’t up to the task of further analysis, so I decided to waste some time making an mpeg or two of the Mandelbrot set. I modified an old .signature program I had to dump a whole bunch of pgm files, and then used a program I call ppmtocmap to extract interesting colormaps from photographs to color them. A few more command line utilities, and voila, an MPEG1 and MPEG2 zoom into the Mandelbrot set.

I’ve been interested in using digital cameras to capture panoramas, and
yesterday I had my cheapy $50 digital camera with me, so I shot 19 handheld shots that whirled around me as I stood outside my place of business. This is the result:

I deleted the applet, it made loading my main page too slow. If you want, you can download your own copy of the Panaview applet and set this up yourself.

If you don’t have java, you can just look at the panorama by clicking on
the tiny thumbnail below:

I’m fairly happy with Movable Type, but I am constantly on the lookout for new weblogging software. During one of my surf sessions, I ran accross blosxom, a very simple weblog written in seven pages of perl. It’s a very cute system, and has some terrific ideas.

Inspired by its brevity and power, I’ve set up the beginnings of a similar system written in Python that I am calling Pylon. It’s only 140 lines of Python so far, but I was able to write a simple Python script to import all of the entries from brainwagon into it.
Check it out.

Eventually the code will be available here under my .signature license.

I’ve scanned and created another document, an article from the Gleanings for the ATM column that ran for years in the magazine Sky and Telescope. This article is by Dick Buchroeder and discusses the plans for a Catadioptric Herschelian Telescope, an interesting telescope design with tilted optics.

If you aren’t interested in telescope making, you might be interested in looking at the quality of the output images anyway. For the majority of the text (all but the first cover page), I just used the Any2Djvu server, but I goofed around a bit with the cover image (the color scan of the cover for the Sky & Telescope). I took the color scan, converted it to a grayscale map, and then thresholded it to produce a binary image that contained the text and some cruft from the cover image of the two watches. If you try to use the cjb2 compressor on this image, you don’t get very good results, but if you paint out the crud, and then use the djvumake program, you can get a very nice image, and one that is entirely comparable in size to the output of the dedicated server.

I’ll probably write up a simple guide on how to do this later, and see if I can
get it posted someplace where more people will see it and hopefully use it.

I did a bit more research on djvu, and have done some more experiments. Long ago, I rescued a huge number of old Scientific American magazines from the dumpster of our local library. I grew up reading the Amateur Scientist column conducted by C. L. Stong, and of course Martin Gardner’s superb Mathematical Games. I ended up redonating most of those magazines to another library (who probably prompty threw them out).

Nevertheless, I do still have a few that include topics of interest, and decided to use one of them as source material for my experiments. In November 1981, the topic was pinhole cameras. I disassembled my copy of Scientific American and fed it to my $50 scanner, generating six 25 megabyte scans. I then converted these to reasonable quality JPEG images, and uploaded them to the
Any2DJVU service that is available on djvuzone. I took the individual pages and downloaded them, and using the djvm program, assembled them into a single downloadable document, which you can see here.

The resulting compression is 550:1. Not too shabby.

This compression seems to be hard to achieve using purely free tools on my machine. In particular, the free tools don’t include a tool for performing the segmentation that makes the overall process tractable. I’ve been toying around with crudely doing the segmentation by hand, but the results aren’t as good as the automatic version.

Anyway, I may convert some other of my available documents into this format
for redistribution, including some hard to find and useful telescope information.
Stay tuned.

After our day of enjoying fine art, we decided to go out and get some delicious Thai food from Thai Spice which is located just south of Golden Gate Park, and then went to the Marin Headlands for a bit of a drive near Point Bonito. The remnants of many old gun emplacements are still visible here. I saw some views of the Golden Gate that I hadn’t seen before, and snapped a bunch of pictures with my cheapy $50 digital camera. When I got home, I stitched them together to make the panorama above. Not the highest quality, but a pleasant view nonetheless.

I have a bunch of papers that are not valuable, but are rare and hard to find. Some I have only as faded Xeroxes, such as a copy of Anton Kutter’s treatise on Schiefspiegler telescopes, and Arnold Leonard’s work on Yolo telescopes. Others are in collections which are out of print, such as Tom Duff’s Polygon scan conversion by exact convolution. I’ve recently decided that I should scan papers that I think are interesting and preserve them in some digital form.

At the Hacker’s conference, Brewster Kahle touted the DjVu format as an alternative to pdf. For fun I decided to try to use my available tools to try to convert my bad xerox of Tom’s paper (15 pages) into a compact, reasonable form. I slopped all 15 pages into my super budget Canon scanner, and quickly converted them into 300dpi bilevel TIFF files. Each file was an 1,053,264 byte file.

The DjVu tools are open source, I got them by installing the djvulibre package in FreeBSD. The program cjb2 provides bilevel image compression of PBM files, so I made a little script that converted each page into a pbm, and then to a .djvu file. I specified that the compression could be lossy and that it should remove flecks, and then assembled them together using the djvm program.

The resulting file was 279,530, for all 15 pages.

That wasn’t quite good enough though, I decided to go ahead and use the online any2djvu server to perform OCR on the djvu file and stash it back inside. The resulting file with OCR is 313,192 bytes, and can be searched.

I then tried to make a pdf file out of it. I converted the djvu file into PostScript (using the djvups program) and then used ghostscript to convert it to a pdf file. The result wasn’t pretty: the PostScript file was 4,268,709 bytes and the resulting PDF file was 3,184,680 bytes, a 10 increase over the DjVu file. I have no doubt that Adobe Distiller could do a better job, but then, I don’t have Adobe Distiller.

Anyway, I thought it was a fun experiment. You can have a peek at the resulting DjVu file if you like. You can either install the viewer or if you have a Windows machine, install the plugin.