GroupXIV is a microphotography viewer based on Leaflet. It allows to cut an arbitrarily large image into tiles, conveniently display them on a desktop or mobile browser, create persistent URLs, and measure distances and areas.

For example, see Atmel ATmega8, Atmel ATtiny24V, Epson S1D15719.

The tile cutter depends on Python 3 and wand.

The viewer has no server-side code, so it will work with any webserver.

First, generate the JavaScript and CSS bundle:

make bundle

Serve the public_html folder from any convenient URL.

For example, you could use Python's builtin HTTP server:

cd public_html && python3 -m http.server 8000

1. Assuming an image called image.png, put image.png in public_html/data.
2. Run python -m tile_cutter public_html/data/image.png from the repository root. This will create:
   • public_html/data/image.png-tiles, containing the sliced image;
   • public_html/data/image.png.json, containing the image metadata.
3. Change the following metadata fields:
   • name to describe the source of the image, e.g. "Atmel ATmega8";
   • scale to contain the ratio of pixels to nanometers, e.g. 540 for 540 nanometers per pixel.
4. Assuming public_html is served from https://groupxiv/, navigate to https://groupxiv/#url=data/image.png.json.

The original image.png is no longer necessary, however it is recommended to keep it for anyone who would like to download the source of the tileset.

You may find that large images fail to convert, giving one of these messages:

wand.exceptions.ImageError: width or height exceeds limit `' @ error/cache.c/OpenPixelCache/3909
wand.exceptions.CacheError: cache resources exhausted `' @ error/cache.c/OpenPixelCache/4095
wand.exceptions.WandError: wand contains no images `MagickWand-1' @ error/magick-image.c/MagickGetImageWidth/6336

These errors are caused by a security policy which has been introduced by ImageMagick to deal with buffer overflow attacks.

To fix this, edit /etc/ImageMagick-6/policy.xml:

• width or height exceeds limit: increase domain="resource" name="width" and name="height" to be greater than the input image.
• wand contains no images: increase domain="resource" name="width" and name="height" to be greater than the tiled size groupXIV prints out.
• cache resources exhausted: increase domain="resource" name="area" (maximum image cache area in pixels) and domain="resource" name="disk" (maximum disk space to use for caching large images).

The following settings seem to work for images up to about 30,000 pixels square, which produced a 32768-pixel tiled square:

• width and height: 70KP (70,000 pixels)
• area: 128GP (128 gigapixels)
• disk: 30GiB (30 gigabytes)

The command identify -list resource can be used to display the current resource limits.

As soon as I have a setup for capturing multi-layer imagery, I plan to add multi-layer support. The JSON metadata format already supports it somewhat.

GroupXIV uses two Leaflet controls developed specifically for it: Leaflet.Nanoscale and Leaflet.Nanomeasure.

• Image stitching software can mitigate a reasonable amount of out-of-focus pixels; even 30% usually produces tolerable results as long as every area is in focus at least once.
• If your imaging setup consistently produces out-of-focus pixels in the same regions, it's best to cut them out, e.g. using ImageMagick: for i in raw*.png; do convert $i cropped-$i -crop WxH+X+Y.
• Image stitching software can mitigate a substantial difference in exposure, but it is instead recommended to keep exposure constant during capture. A good idea is to find an area where a low-reflectivity area, such as many thin metal interconnect traces, is immediately adjacent to a high-reflectivity area, such as a metal polygon, and adjust exposure so that neither is under- or overexposed.

Hugin is a very powerful application for stitching images, however its intended domain is panoramas and the UI does not make it easy to stitch flat tiles. Here is a step-by-step guide:

1. Select Interface → Expert.
2. On Photos tab under Lens type, select Add images.... When prompted for field of view, enter 10; this value is not important.
3. On Photos tab under Feature Matching, Settings:, select "Cpfind (multirow/stacked)". Click Create control points. This can take a few minutes to a few hours.
4. On Photos tab under Optimize, Geometric:, select "Custom parameters". Do not click Calculate yet.
5. On Optimizer tab (tab appears after step 3) under Image Orientation, right-click on every column except TrX and TrY and click Unselect all. Right-click on TrX and TrY and click Select all. Make sure only values, and all values, in TrX and TrY columns are bold. (If your images are not level, add Roll to that set.)
6. On Optimizer tab (tab appears after step 3) under Lens Parameters, right-click on every column and click Unselect all.
7. On Optimizer tab, click Reset, select every checkbox and click OK.
8. On Optimizer tab, click Optimize now!. This can take a few minutes to few hours.
9. On Photos tab, select all photos, right-click, select Control points, Clean control points. This can take a few minutes.
10. On Optimizer tab, click Optimize now! (again). This can take a few minutes to few hours, but quicker than the first one.
11. Select View → Fast Preview window. This will open a new window.
12. In preview window, under Projection tab, left list box, select "Normal (rectilinear)".
13. In preview window, use the sliders to the bottom and the right to fit the image in the viewing area; under Move/Drag tab, Drag mode:, select "mosaic", then draw the image at the center. It allows you to estimate whether the fit is good. A good fit is seamless and all straight lines on the sample should appear completely straight in Hugin.
14. In preview window, select the Crop tab, then move the areas that are highlighted when you move the cursor near the edges of the viewing area so that only the sample is inside the white rectangle.
15. In main window, under Stitcher tab, under Canvas size:, click Calculate optimal size; under Panorama Outputs: select "Exposure corrected, low dynamic range"; under Panorama Outputs:, Format: select "PNG", under Processing, Blender:, click Options, enter --fine-mask, click OK.
16. In main window, under Stitcher tab click Stitch!. This will first open a save dialog for the Hugin project, then it will open another save dialog for the panorama output as well as intermediates (which will be temporarily placed in the same location as the panorama output), then it will open a PTBatcherGUI window. PTBatcherGUI could complain about assertion failures; ignore that.
17. PTBatcherGUI will automatically process all files in a few dozens of minutes to a few hours. Done!

Key points:

• Multirow CPfind is an optimal control point search method for images that are taken sequentially in multiple rows, taking less comparisons than a generic pairwise method.
• Microphotography has practically no optical distortion to speak of and the images are usually perfectly, or near-perfectly level. Thus the only parameters Hugin should try to adjust is the X and Y translation. If it tries to adjust others, it will certainly overoptimize, especially on highly similar images such as large chunks of metal interconnect.
• For the same reason, Hugin should assume a rectilinear lens, i.e. a lens that makes straight lines appear straight on the pictures.
• --fine-mask is a workaround for a rare but annoying enblend bug.
• Trying to move images around in any way except with a rectilinear lens and mosaic mode will change positional parameters of the images and it'll be necessary to reset these and re-optimize.

Alternatively, replace the steps 1-13 with the following script:

#!/bin/sh
set -e

PROJECT="$1"
IMAGES="$2"

if [ -z "${IMAGES}" ]; then
  echo >&2 "Usage: $0 <projectname> '<image_glob>'"
  echo >&2 "  e.g.: $0 attiny2313 'raw*.png'"
  exit 1
fi

pto_gen -o ${PROJECT}_1.pto ${IMAGES}
cpfind -o ${PROJECT}_2.pto ${PROJECT}_1.pto --multirow
pto_var -o ${PROJECT}_3.pto ${PROJECT}_2.pto --opt TrX,TrY
autooptimiser -o ${PROJECT}_4.pto ${PROJECT}_3.pto -n
cpclean -o ${PROJECT}_5.pto ${PROJECT}_4.pto
autooptimiser -o ${PROJECT}_6.pto ${PROJECT}_5.pto -n
sed >${PROJECT}_7.pto <${PROJECT}_6.pto -e 's,^p .*$,p f0 w3000 h1500 v179  E0 R0 n"TIFF_m c:LZW r:CROP",'

From that you still need to proceed via hugin's GUI.

MIT license