IEEE Systems, Man and Cybernetics Magazine - July 2019 - 62

repeat-pass coregistration, creation of a change map,
detection of changes, and reduction of false alarms (false
detection). Figure 2 is a block diagram of ACD processing. The process begins with a historical database comprising SAS images with their associated geographical
locations. This database provides the baseline or reference data. Given a repeat-pass image, the database identifies an image corresponding to the same geographical
location and is used in the following coregistration,
change-map generation, change-detection, and falsedetection reduction steps.
Coregistration
Image coregistration is a key step in the ACD process. For
ACD to successfully process two corresponding images,
they must be coregistered accurately. With SAS imagery,
the difficulty in coregistration is due to challenges such as
vehicle trajectory, low resolution, the presence of noise,
changing features of the sea floor, and biologics in the
water column. A three-stage process is utilized in the ACD
process: navigation alignment, fine-scale coregistration,
and subpixel coregistration. In the first stage, the nominal
path from the starting to the ending location of the reference and repeat-pass images are aligned, allowing for

Input

Change-Map Generation
The change-map generation step provides maps, using tools
such as coherent and incoherent change detection (CCD and
ICD, respectively), that discern changes in the ocean bed
[20]. This step relies heavily on the previous coregistration
step. If the two SAS images are not coregistered correctly,
the resulting change map will not be reliable. The CCD tool
generates a coherence map by predominantly exploiting the
phase signal from the coregistered multitemporal imagery
to discern the presence of new sea-bed features. Due to the
dependence on the data being coherent and the dynamic

Change-Map
Generation

Three-Stage Coregistration

Reference

Repeat
Pass

repeat-pass data to be projected onto the corresponding
reference image. Next, in the fine-scale coregistration, the
scale-invariant feature transform (SIFT) algorithm [24],
[25] is applied to match features in the reference and
repeat-pass images. Finally, in the subpixel coregistration
stage, coherence corrects any navigational errors not
addressed in the first two stages. Here, the two images are
locally coregistered and their underlying phase data are
coregistered. An example of a coregistered pair of images
is presented in Figure 3(a) and (b). A detailed description
of the SAS image coregistration is conveyed in previous
work [20].

Fine-Scale
Coregistration
(SIFT)

Subpixel
Coregistration

Detection

Data
Reduction

Change-Map
Detection

Component
Analysis

CCD

ICD
Navigation Alignment

Figure 2. A general block diagram of an end-to-end ACD process.

(a)

(b)

(c)

Figure 3. The coregistered (a) initial pass and (b) repeat pass images of an area and (c) their ICD change

map. The object in the initial pass was moved and a new object is present on the right side of the repeat-pass
image.

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE Ju ly 2019

IEEE Systems, Man and Cybernetics Magazine - July 2019

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - July 2019