IEEE Geoscience and Remote Sensing Magazine - March 2013 - 45

I. INTRODUCTION
s the amount of NASA data is increasing in all Science domains (Earth, Astrophysics, Heliophyics,
Planetary) as well as in the Exploration domain, the need
for automatic or semi-automatic and accurate image processing techniques tailored to NASA datasets is increasing. In some cases, for example for disaster management
and monitoring or in Exploration, these techniques also
require execution (or results) in real- or near-real time, as
their outcome is the driver of human or autonomous decision systems. State-of-the-art image processing technologies have been developed and applied for many years to
the military, medical and commercial domains, but the
application of these methods to NASA image processing
challenges lags behind these other domains. Often, the
barrier to test newly developed algorithms on NASAgenerated data is the difficulty to easily obtain image data
in a format that is manageable by the university computer
science community, as well as a lack of knowledge of the
application domain and of the desired scientific results.
As a consequence, the application of image processing to
NASA data has often been restricted to older, well-tested
off-the-shelf methods that are not always optimal for
NASA applications.
An example of this last issue is illustrated by the
problem of geo-locating remotely sensed image data [1].
The geo-location problem can be summarized in the
following way: when an image is acquired by a satellite
instrument, it is relatively easy to align the image with a
specific geographic location in a general sense (e.g., "this
is an image of the Chesapeake Bay"). However, meaningful scientific results require very accurate determination of the images geographic location (or geo-location).
A good geo-location (or registration) algorithm has to
take into account many factors such as satellite orbits,
sensor characteristics, atmospheric and cloud interactions, multi-temporal and multi-sensor effects, etc. After
a systematic geometric correction based on the sensor
specific navigation model is performed, a more accurate
geo-location can be obtained by using image registration based on image features. Until very recently, most
remotely sensed pairs of images were registered in systems such as ENVI, using semi-automatic methods, with
a very labor-intensive first step of visually locating corresponding features in the pair of images. It is only in the
last few years that many projects (e.g., Landsat, MODIS,
SeaWIFS, MISR) have developed image registration
methods to perform this step entirely automatically [1].
But these methods are adopted on a case-by-case basis
and the tools currently included in off-the-shelf packages do not generally satisfy all Science projects requirements. However, such registration methods have existed
in other domains for many years, for example in the
medical domain where benchmark data including fiducial points are available to validate new image registration methods [2].

march 2013

ieee Geoscience and remote sensing magazine

Generally, our goal is to assemble a collection of datasets with applications to various NASA domains such as
Climate Modeling, Planetary Exploration, and Applied
Sciences (e.g., disaster management, water resources and
invasive species). When building IMAGESEER, the focus is
on addressing the specific NASA challenges, namely:
◗◗ The large variety of sensors and acquisition conditions
◗◗ The large size of datasets and their NASA-specific formats
◗◗ The lack of a reference image model (unlike the medical domain)
◗◗ The knowledge of the application domain, known from
Image datasets have
NASA scientists and their
been selected to
collaborators, but often unknown to computer scientists
represent specific
◗◗ The lack of ground truth or
NASA challenges
"truth data" to facilitate the
dealing with data
validation of new algorithms.
size, complexity
This paper will first provide
and semantics.
a brief survey of other image
databases available to the Image
Processing community and will
describe the specific needs that led to this NASA-centric image
database. Then, the technical description of IMAGESEER, its
contents and its web-based user interface will be given with a
few illustrative screen shots. Finally, future extensions of this
preliminary prototype will be described.

II. Need for a Nasa-Centric Image Database
A. Objective
The overall objective of IMAGESEER is to design a database
of NASA benchmark image data to:
1) Enable teaching Image Processing techniques on illustrative NASA-centric applications data
2) Provide benchmark NASA Science data for new research algorithms
3) Foster Image Processing collaborations between NASA
and university/industry/international research groups.
B. Need for a NASA-centric database
As briefly listed in the Introduction, the specific challenges
associated with the processing of NASA image data include
the following:
◗◗ The variety in the types of sensor data and the conditions of
data acquisition. A technique that appears to work accurately on satellite imagery acquired at a given time
over some given location may not perform as well on
data from the same sensor at other times or over another location. For example, in Earth Science, varying
atmospheric conditions (e.g., clouds, sun angle, etc) can
distort sensor readings taken of the same location at different times.
◗◗ The size of the data. For example, a typical Landsat scene
is of size of about 7000x7000 pixels on average, contain45

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2013