Image Content Engine (ICE): A System for Fast Image Database Searches Page: 4 of 7
This article is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided to Digital Library by the UNT Libraries Government Documents Department.
The following text was automatically extracted from the image on this page using optical character recognition software:
The ICE search interfaces allow the user to interactively select target matches from a set of thumbnail images defined
by peaks in the object matching metric. Search targets can also be specified by sets of example images. The query
interface returns image tiles with feature distributions similar to the query examples. The use of both the model-based
search and the query-by-example interfaces will be described in more detail in Sections 3 and 4.
Object matches that are confirmed by the user can be stored in an object database. These objects and their
characteristics can be utilized by higher-level semantic graph-based knowledge management systems that correlate data
from multiple sources.
A GIS interface is being developed for ICE which will allow the results of searches to be displayed as layers in a GIS.
This capability will allow image information from multiple sources to be viewed in a common geospatial coordinate
system and allow correlation between image objects detected by ICE and a priori map information. The search strategy
in ICE will also be controllable by the GIS interface, allowing us to search specific map-specified areas. For example a
search for a specific class of vehicles can be set up to follow known roads on a map.
The heavy computational loads in ICE reside in the image processing pipelines. To implement these pipelines
efficiently we have developed a set of software tools for mapping arbitrary pipeline operations on Linux cluster-based
computer architectures. Image blocks are allocated to sets of processors by a master control process. Image metadata is
managed by passing XIIL descriptors with the image data blocks using the ISP toolkit. We are testing the scalability of
the system using the MCR parallel system at LLNL. Most operations to date have been on small subsets of MCR's 989
nodes but the data-parallel approach should allow scaling to large parallel systems.
Engine Pipelines -
Query by I
Object Matching Emp 1cle a~ns
Engine Pipelines MIce Object ,P
Database User-Interactive IDatabase L6
Future IE ICE Search Interaces
ICE Processing Engines
GIS Interface Layer
Figure 1: Architecture of the ICE image search system.
3. MODEL-BASED TARGET SEARCH
The principal function of ICE is to search large image sets for objects defined by a three-dimensional target model. The
target model is given as a 3D graphics model file. It is projected into image space by a sensor model which captures the
imaging geometry, illumination, and any other system aspects that influence the resulting image. Current models are
relatively simple but interfaces to arbitrarily complex models are in place. The sensor model must adapt its performance
to the quality of the image being searched. If the image is low resolution too many details in the model can degrade
Here’s what’s next.
This article can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Article.
Brase, J M; Paglieroni, D W; Weinert, G F; Grant, C W; Lopez, A S & Nikolaev, S. Image Content Engine (ICE): A System for Fast Image Database Searches, article, March 22, 2005; Livermore, California. (digital.library.unt.edu/ark:/67531/metadc887962/m1/4/: accessed November 19, 2018), University of North Texas Libraries, Digital Library, digital.library.unt.edu; crediting UNT Libraries Government Documents Department.