|
Overview
SARscene™ is an image analysis environment with special focus on context mapping and vehicle detection tools.
Although designed for synthetic aperture radar imagery, the toolset has broad application to a variety of
image modalities, and can be used for general image processing tasks, such as viewing, segmentation, and the
development of ground truth. SARscene™ includes advanced SAR target detection and context mapping tools
developed by SET researchers. Based on subaperture phase coherence properties, our target detection tool
provides high detection rates with few false alarms, and is highly effective for close proximity targets.
Our semi-autonomous context-mapping tool allows users to quickly segment SAR imagery to aid in the reduction
of false alarms.
Features
-
Segmentation - SARsceneTM has a variety of configurable segmentation algorithms used to develop
context maps, including trained and untrained algorithms. Trained algorithms require the additional
input of sample image data for each label, while untrained algorithms do not require such data. For
those that require trained data, manual and automated (using DFAD data) training data collection
methods are available.
-
Vehicle detection - Three configurable vehicle detection algorithms are available, including subaperture
coherence, CFAR, and polarimetric phase. In SARsceneTM's context based FA mitigation, a context map can
also be used in conjunction with the results of vehicle detection to reduce the number false alarms.
Depending on the algorithm, vector images (images that have a vector associated with each pixel, such as
extracted texture features), complex images, as well as images with float and integer data can be used as
input for segmentation.
-
Image processing - SARsceneTM also has several image operations that may be used as post processing for
either context maps or vehicle detections. They range from down sampling and median filtering to n-level
operations, which performs morphological operations on a single label of a multi-label image. It also has
the capability for feature extraction, which creates vector images that may be used as input for segmentation.
-
Image viewing - SARsceneTM is able to display vector, complex, float, and integer images in a variety of ways.
For example, images may be rescaled, filtered, zoomed in, and rotated. Each image of a vector image, and the
magnitude and phase of a complex image can be viewed. A label image may also be colorized to analyze the results
of segmentation or vehicle detection. Highlighting is available to combine the colorized results and the input
image for easier analysis.
-
Construct site models for ground truth inspection - SARsceneTM includes an application featuring manual
ground truth generation, image overlap comparison, and histogram data analysis. In constructing site models,
it allows the user to manually draw site objects using closed shapes. These include rectangles, polygons, and
ellipses. Each of the drawn site objects is associated with a site label. Site labels are configurable, enabling
the user to change the site label properties, such as its bitmap level for masking and the color of the sites
drawn onscreen. Additional site labels may also be added, or removed. Drawn site objects can be stored into a
data file, and later loaded to continue adding new site objects or to edit site objects already developed.
These site objects can ultimately be used to produce a label image for ground truth evaluation or as a manually
generated context map.
Requirements
- 500 MHz PC running Windows NT 4, Windows 2000, or Windows XP.
- 128MB of available RAM or greater
- 15MB free hard drive space
- 16MB video card (1024x768 resolution or greater)
- Java 2 Runtime Environment (JRE) 1.4.1 [Download]
- Adobe Acrobat Reader (to view manual) [Download]
Downloads
Previous Versions
Installation Procedures
- Please see the README.txt file included in the package.
Support
|
