IEEE Geoscience and Remote Sensing Magazine - December 2021 - 121

(a)
(b)
(c)
FIGURE 1. Multitemporal optical image geometrical dislocation. (a) A reference image taken by Landsat 5 on 15 October 1990. (b) A sensed
image taken by Landsat 5 on 15 September 1993. (c) The overlapping images of (a) and (b).
images covering the same scene. Thus, geometrical registration
techniques are implemented to align two or more
images from the image-to-image perspective rather than
the imaging mechanism. Consequently, geometrical registration
is an image-processing technique that aligns different
images of the same scene acquired at various times
and viewing angles and with multiple sensors [1]. As a fundamental
task in remote sensing information processing,
it is a prerequisite for many practical applications, such as
image mosaicking [2], image fusion [3], land cover change
detection [4], [5], and disaster evaluation [6], [7].
It worth noting that there is a technical term, coregistration,
that is similar but not exactly the same as image registration.
It is now commonly used in aerial and unmanned aerial
vehicle image registration, generally including multimode
registration and alignment through the aid of auxiliary data.
When the registration is conducted with a GPS/inertial measurement
unit, it usually establishes a connection between
an image and the simulated or real ground [8]. Certainly,
the registration technology works on tie points generation
for the construction of relationships. With real ground control
points (GCPs), the tie points between the reference and
sensed bands are produced to register different bands of hyperspectral
images [9]. Additionally, when the orientation of
the reference image is determined, without GCPs, the coregistration
of multitemporal high-resolution image blocks is automatically
achieved [10]. Although there are time-increasing
papers focused on coregistration techniques doing some auxiliary
work with the positioning data, the core of the process
is image registration, as far as we are concerned. Therefore,
the emphasis is put on the opportunities and challenges of
geometrical registration in remote sensing fields.
Geometrical registration can be traced to the 1970s,
when the United States proposed image registration to analyze
target objects in aircraft-aided navigation and weapons
systems. Since then, it has rapidly developed, particularly
in the domains of remote sensing, computer vision,
DECEMBER 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
and medical image processing. Some conclusive studies of
computer vision and medical image processing have been
published [11]-[16]. Building on a widespread survey of image
registration, published in 1992 by Brown [15], a 2003
review [16] comprehensively summarized the subsequent
research. In recent years, several overviews of image registration
have focused on newly developed approaches inspired
by extant versions [17]-[19]. However, these surveys
are limited to analyzing and drawing conclusions based on
conventional approaches [20]-[22]. Since the first study
of multispectral and multitemporal digital imagery registration
in 1970 [23], an increasing number of papers have
contributed to the field. A total of 140,983 related studies
with the keywords image registration or image matching
were retrieved, from 1979 to January 2021, from Web of
Science (WoS). When screening again using the keyword
remote sensing, 46,141 articles were found, as plotted, based
on their publication year, in Figure 3. The respective proportions
of the total number of papers on WoS per year are
also presented. It can be seen that a small number of papers
T
T + t
FIGURE 2. The angle difference from multitemporal images in a
mountainous region.
121
IEEE Geoscience and Remote Sensing Magazine - December 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2021
