required by the system results in a significant reduction in the
overall cost of the system. At first blush it might seem that the
cost associated with processing is higher when images are stored
in a compressed format; however, this need not be the case. If
the data are organized and stored in the appropriate fashion, the
search process can be made more efficient and we do not need to
analyze all ofthe information stored in the image in order to
eliminate it from consideration. One of the primary ideas of our
project is that it is possible to increase the speed of searching
through images stored in a digital library while simultaneously
reducing the storage requirements. The remainder of this section
goes into more details on this topic.
Compression techniques can be either “lossless” or “lossy”. A
lossless compression scheme is one which guarantees perfect
reconstruction of all of the bits in the original dataset. A lossy
compression scheme, on the other hand, does not reconstruct
most images exactly but rather allows the loss of some informa-
tion in order to achieve higher compression ratios. The end-user
requirements determine whether lossy or lossless compression
schemes can be used.
As discussed below, our analysis scheme takes advantage of the
properties introduced by lossy compression. However, remote
sense scientists are reluctant to lose any data and, therefore,
demand lossless compression. The compression scheme we use
for image storage offers a hybrid solution. In order to address
this issue, we employ a wavelet-based scheme that allows us to
progressively extract image content by specifying both spatial
and spectral constraints. Typically, as we relax the spectral con-
straints we tighten the spatial constraints. However, if one fully
relaxes both constraints the output is a lossless representation of
the original image. The overall compression ratio achieved by
this scheme is competitive with the best lossless compression
schemes we have analyzed to date.
In general, applying query and retrieval operations directly on
lossily compressed data leads to improved computational effi-
ciencies along two fronts:
» One needs to process fewer bits;
» The features and properties of the data are emphasized by
the transformed-based compression.
Query operations including retrieval, evaluation, transmission
and visualization of the image (or video) data can be staged pro-
gressively, by selectively and adaptively processing limited
amounts of information, to minimize the total execution time.
The difficulties that get introduced are twofold:
* As the number of coefficients is reduced, dimensionality of
the search space is reduced, resulting in many-to-one map-
ping. Thus, the number of false hits increases.
* The reduction of coefficients implied by the compression
results in alignment errors. The net result is that the output
produced by the operation may not match the exact value
produced by operating directly on the original image.
* We have developed techniques that, for several operations,
allow us to quantify the latter effect without having to con-
vert back to the spatial domain. This allows us to guarantee
194
that the results of our filters are identical to those produced
by operating directly on the full image.
Access via Internet
Access to and dissemination of the data in the INFER project is
provided via the Internet so data management techniques which
improve access, search, and retrieval time, interactive visualiza-
tion techniques, and a Netscape-based user interface are all inte-
gral components of the system. While each of these components
could be discussed at length, this paper will focus on only one -
that of the need for a more complex, or ‘smart’ user interface in
an Internet context.
Through development of the INFER prototype it has become
clear that an extremely critical component of an internet applica-
tion is the user interface. In a local software system, the user
interface must be intuitive enough for a user to navigate easily
through the system functionality but, in general, a user working
with in-house applications has an understanding of the underly-
ing data.
As we move to Internet-based applications it is important to con-
sider the fact that the users are, in effect, logging on to a black
box, with little or no knowledge of either system functionality or
server databases. In the case of providing a data archive search
capability it is essential that the user interface assist the user in
understanding both the search functions and the extent of the
data available for search. In effect, the result must be a ‘smart’
user interface.
In a simple browse of a large data archive this means supplying
the user with information on the spatial and temporal extent of
the data. In an application scenario the problem becomes much
more complex. For instance, in the context of a content-based
search the user must be able to understand and specify the fol-
lowing individual components of the search and their interrela-
tionships:
. Possible search features or search tools - this can be
equated to the ‘fields’ in a database.
° The data sets available for search, e.g. AVHRR, DMSP.
° The temporal extent of the data sets.
° The spatial extent of the data sets.
Providing this information to the user can present a user inter-
face problem as user selections will progressively change the
nature ofthe available data. For instance, specific search features
may only be supported in a subset of the data; by selecting to
search on that feature the user must be notified of the potential
spatial and temporal extent of the search. On the other hand,
beginning a search by selecting a spatial area may constrain the
user to a subset of the search features. Essentially, wherever the
user starts the search the information relative to the other three
components may change and notification is required.
The user would also be interested in knowing what ancillary data
sets are available for background information or for complex
searches. For instance, the user might want to view the results of
a search for fires relative to topography or may want to actually
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
con
gra
Uni
and
infc
fac:
opti
ofs
but
Pro
dep
One
cod
latic
Thi
nati
bas
For
seat
pot
rest
forr
INF
IBN
exaı
wilc
rese
date
cati
ing
pop
able
resp
tion
ing |
Wil
bilit
and
bou
atio!
patc
to fz
ther
nati
arch
fire
mat
mul
data
arie:
mod
earlı
fire |
