Image Content Engine (ICE): A System for Fast Image Database Searches Page: 5 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:
Matching the projected model to the image is performed by a phase-sensitive detection algorithm. We compute the local
image gradient or image flow and match the phase angle off the gradient with that of the model. This method reduces
sensitivity to illumination changes. The algorithm is described in detail in .
is - 'i
x- rr r
t ;, _ -y
". "V'4 " w
r ' Vi
Current File C \ICEimageslRussia\03U1L22082512-P243 R10C1-000000076333_01 PO0l NTF
Thumbnail 1 of 7588
Thumbnail Figure of Mert 0 821920
Focus Point (col, row) ( 11571, 1332)
t r ,r
Thumbnails 1 thru 8 of 7588
Figure 2: The user interface for selecting detection thumbnails from the model-based search process (Images courtesy of Digital
Figure 2 shows an example of the user interface for model-based search. In this case the model described the distinctive
building shown in the center of the image. The thumbnails on the right show multiple detections of this class of building
in the image database. The thumbnail positions are determined by locating peaks in the matching metric surfaces.
Clicking on any of the thumbnails brings up a wider view of that specific image in the left window. The user can scroll
down through the set of returned thumbnails and mark those that are acceptable detections.
4. SEARCH BASED ON EXAMPLES
An alternative approach to specifying a search is through the use of example images [2,3,4]. Each image tile - an
arbitrary rectangular set of pixels - is described by a hierarchical set of image features, the feature vector. If we choose
an example region and compute its feature vector, we can then search the index of feature vectors for the full image
database and return those that are in some sense closest to the query feature vector.
Our approach to computing a feature vector is shown in Figure 3. The image is represented by a hierarchical set of
feature maps. At the first level we compute a 5-scale Laplacian pyramid  representing local image contrast at each
scale. A thresholding saliency operator selects features in the maps to be passed to the higher levels. The saliency
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. (https://digital.library.unt.edu/ark:/67531/metadc887962/m1/5/: accessed March 26, 2019), University of North Texas Libraries, Digital Library, https://digital.library.unt.edu; crediting UNT Libraries Government Documents Department.