409
FACTOR ANALYTIC TECHNIQUES FOR ENHANCEMENT
OF TEMPORAL INFORMATION IN DIGITAL IMAGERY
Fredrick C. Luce
Senior Research Technologist
Brian J. Turner
Associate Professor of Forest Management
Co-Director, Office for Remote Sensing of Earth Resources
Office for Remote Sensing of Earth Resources
Institute for Research on Land and Water Resources
The Pennsylvania State University
University Park, PA 16802
ABSTRACT
Under contract with the Defense Mapping Agency, four digital processing
techniques for detecting change in digitized aerial photos using the
ORSER software system were evaluated. These four techniques are as
follows: 1) digital post-classification comparison of classified scenes
from two dates; 2) classifying the two date-image data as one data set;
3) density-slicing ratios and differences between the two data sets;
and 4) finding a transformation to highlight temporal information using
factor analysis. All four techniques were applied to an initial test
site and the factor analytic technique was shown to be the most success
ful for mapping temporal information. This technique was then applied
to four additional test sites and further evaluated. Initially, five
factor rotations were used on each test site, using all the principal
components. The best transformation was selected, density sliced, and
displayed to highlight the temporal information. A similar analysis
was then applied to the first few principal components. This, however,
did not improve the enhancement of the temporal information.
1. INTRODUCTION
An efficient method of analyzing land-use change over time involves the
use of remote sensing techniques. In the past, detection of change was
accomplished by visual interpretation of photographs. The photointer-
preted classification results from an area at some initial time were
compared (manually and visually) to the classification of the area at a
later date (Theis, 1979). Another visual method was the "blink" pro
cess. This involved viewing the two images in a rapid or blinking
fashion (Masry et al., 1975). This process is both tiring for the
operator and insensitive to subtle image differences. With large areas
of multiple-date imagery, manual analysis of temporal information has
been a time-consuming and labor-intensive task. Although these methods
have been useful and accurate, they have not taken advantage of comput
erized techniques for data handling and storage.
While digital analysis of imagery has allowed more efficient analysis of
temporal information, particularly over large areas and several dates of
data collection, the techniques used in the analysis have essentially
remained the same. That is, a digital classification from one date has
been compared (visually, manually, and/or digitally) to a digital classi
fication of the same area at a second date. The numerical nature of
digital data, however, lends itself to new methods for highlighting
temporal information. These methods include image differencing and