Loading
Research Projects
Large-scale Image Stitching and De-ghosting
|
Abstract
Stitching a set of partially overlapping images together to form a composite image has received
a lot of attention in various contexts in the last fifteen years. Here, we are interested in
image stitching as a tool to obtain very-high-resolution composite images (or "mosaics") over an
extended field of view, and document a prototype pipeline to effect high-quality, ghost-free mosaics,
while approaching real-time performance. By exploiting the configuration geometry of the
image acquisition process, our pipeline is able to deal with cases where adjacent images exhibit
very little, noisy overlap. Development of the pipeline has focused around camera arrays, but it
is applicable to any case where an approximate geometry can be obtained, such as when using
cameras mounted on robotic arms, or augmented with motion sensors. Results are presented
for images taken with the AWARE-2 camera array.
Advisor(s): Xiaobai Sun
Carlo Tomasi, Nikos Pitsianis