Slashdot Mirror
Google Unveils 'Gigapixel' Camera To Preserve and Archive Art (thestack.com)
An anonymous reader writes: The Google Institute has developed an ultra-high resolution gigapixel Art Camera which can automatically recompose images into single works of extraordinary detail. The first thousand images are released today, and include works by Rembrandt and Van Gogh. A gigapixel contains over one billion pixels, providing a level of detail unavailable even to the naked eye. The Art Camera has increased the number of available gigapixel art images from 200 to 1000 since 2011. The Art Camera consists of a robot camera that automatically takes hundreds of high resolution close-up photos of the details of an image, using laser and sonar technology to ensure that each image is in focus. Software is then used to take the hundreds of individual close-up pictures and combine them into one whole image. With this technology, one can view photos produced by classical artists from a computer or mobile device without needing to travel around the world to do so. These digital gigapixel images are intended to be available for viewing and studying for years. In the future, we may see Google use machine-learning algorithms to analyze influential classical painters and create new masterpieces.
While the overall result is impressive, the "stitching" isn't perfect. On most pictures it's hard to tell since the brushstrokes have lower resolution than the photography. But on one picture in particular, called "O Livro (os Cem)" by Jac Leirner (1987), the stitching irregularities are easy to find. Type "O Livro" in the search box to find this image.
Essentially this picture is a giant canvas of words in Portuguese. (I speak Portuguese, and it starts off as a bunch of rambling thoughts on money and love, degenerating into what to me makes no sense).
Anyway, to pick an easy to locate spot where stitching apparently took place, find the line about 2/3 down that consists of a giant hexadecimal number (what the hell is that, anyway?). The line starts "D21D22C23..." Blow it up to maximum resolution. The first and second D, and the second 2, have alignment artifacts, and the lower portion of this starting string is slightly blurrier that the top portion. This even gives some insight into the algorithm, where you can see that it's desperately trying to align the top and bottom portions, even distorting or shifting some in-between parts to achieve the result.