|
Image segmentation is a technique used in computer vision for identifying
regions in an image as separate components. The algorithm subdivides
a given image into its constituent regions so that objects of interest
can be identified and analyzed. For example, it is used in the analysis
of MRI images in order to identify structures such as organs or
tumors. Other applications include transportation (traffic monitoring),
defense (target detection), and materials science (fracture analysis).
Because image processing algorithms are application-specific and
frequently tuned, MATLAB® has become a very popular tool in this
domain. Furthermore, a variety of 3rd-party toolkits and extensions
make it a rich platform for image processing.
An image of a cross-section of a human eye (left) and the image's
segmentation (right).
Regions of the original image thought to be part of the same object
are colored with the same color.
|
|
New generations of imaging equipment are driving an ever-growing volume
of images to analyze. As megapixel cameras become mainstream, as 3D
and real-time imaging becomes more commonplace, the computational
workload of image processing sky-rockets.
|
|
Researchers at the Massachusetts Institute of Technology
used the Insight Segmentation and Registration Toolkit (ITK),
an open source image processing library written in C++, in conjunction
with MATLAB®. The goal was to construct an interface to allow parallel
segmentation of images seamlessly from within the MATLAB® environment.
Segmentation with image split into eight vertical slices (left)
and the merged image (right).
|
|
The Star-P Software
Development Kit (SDK) was used to connect the ITK to the
Star-P server. To parallelize the algorithms, the images were stored
as distributed matrices on the parallel server; the watershed segmentation
algorithms were run on each processor's data; and the segmented image
slices were merged back together in MATLAB®.
|
|
 Summary
& Metrics
- MATLAB®-based image segmentation algorithm
- No need to re-write in C/MPI
- Using Star-P server SDK, plugged in
an open-source toolkit to execute in task-parallel mode
- Significant gains in processing speed
(3X speedup on 8-processor server)
 Image
Segmentation
Presentation.pdf >
Image
Segmentation
(PDF) >
|
|
Star-P Support
Blog
Email
Page
