How do you think the new GigE standards will influence the machine vision industry?
Respond or ask your question now!
IMAGERS OF THE YEAR
The Archimedes Palimpsest, a 10th-century manuscript containing the oldest known copies of several treatises of Archimedes, was overwritten in the 12th century with a prayer book. An imaging team has been applying various imaging techniques to enhance the visibility of these writings. Each bifolium of the manuscript is about 300mm x 200mm, and the team collects 10 overlapping images of the manuscript with a 3K x 2K digital camera at about 25 pixels per mm. They subsequently stitch the 10 processed images together to create a digital image of about 7500 x 5000 pixels. The stitching task is difficult because the pages are not flat-the parchment is significantly "cockled" or "wrinkled" and these height variations lead to "blur" in the stitched images.
In Spring/Summer of 2001, Roger Easton investigated the available stitching programs and decided that RealViz Stitcher would do an adequate job. He later tested a future release-versionn 3.0-in the lab and found that it was much faster. Still, later they wanted to test v3.5. While they were still using Stitcher v1.5, a graduate student was also looking into the possibility of writing an algorithm to account for the cockling. They have since switched to version 3.5.
"Our use of Stitcher is, no doubt, slightly different from its original intention," states Easton. "We have three sets of images taken under identical conditions that we want to stitch, so we create a 'stitcher' file for one set, then copy the second set of images to the same directory with the same file names and apply the original 'stitcher' file to the second set, and then repeat again for the third."
The time required is the most significant constraint-it can take about 30 minutes to stitch one set of 10 images. Easton states that they have always wanted to stitch the images at "full resolution" (i.e., one pixel in the stitched image corresponding to one pixel on the original image). In the rendering step they can only approximate this goal now.