ch should be done on
te snow mapping
E
h Observing System
earch was funded by
evelopment Plan of
122004), the special
)2092411) and State
f China (Grant No.
nating clear sky from
asis document.
cloud detection and
ical research letters,
r's guide. University
t. al., 1992. Remote
r properties from the
C JEEE Transactions
|-27.
. Radiative processes
Scientific Publishing
sensing for snow
nces Journal, 41(4),
ng over the Tibetan
Second International
Global Changes,
Society for Optical
EXTRACTION OF BENTHIC COVER INFORMATION FROM VIDEO TOWS AND
PHOTOGRAPHS USING OBJECT-BASED IMAGE ANALYSIS
M. T. L. Estomata* *, A. C. Blanco 5, K. Nadaoka *, E. C. M. Tomoling *
* Environmental Systems Applications of Geomatics Engineering Research Laboratory, Dept. of Geodetic Eng’g, College of Eng'g,
University of the Philippines, Diliman, Quezon City, Philippines — *trixia.estomata@gmail.com, edcarla.tomoling@yahoo.com.ph
® Geodetic Engineering Faculty, College of Eng’g, University of the Philippines, Diliman, QC, Phils. — ayeh75@yahoo.com
° Tokyo Institute of Technology, Ookayama, Meguro, Tokyo — nadaoka@mei.titech.ac jp
Commission VIII, WG VIII/9
KEY WORDS: Marine, Oceans, Mapping, Bathymetry, Recognition, Object, Camera, Photography
ABSTRACT:
Mapping benthic cover in deep waters comprises a very small proportion of studies in the field of research. Majority of benthic cover
mapping makes use of satellite images and usually, classification is carried out only for shallow waters. To map the seafloor in
optically deep waters, underwater videos and photos are needed. Some researchers have applied this method on underwater photos,
but made use of different classification methods such as: Neural Networks, and rapid classification via down sampling. In this study,
accurate bathymetric data obtained using a multi-beam echo sounder (MBES) was attempted to be used as complementary data with
the underwater photographs. Due to the absence of a motion reference unit (MRU), which applies correction to the data gathered by
the MBES, accuracy of the said depth data was compromised. Nevertheless, even with the absence of accurate bathymetric data,
object-based image analysis (OBIA), which used rule sets based on information such as shape, size, area, relative distance, and
spectral information, was still applied. Compared to pixel-based classifications, OBIA was able to classify more specific benthic
cover types other than coral and sand, such as rubble and fish. Through the use of rule sets on area, less than or equal to 700 pixels
for fish and between 700 to 10,000 pixels for rubble, as well as standard deviation values to distinguish texture, fish and rubble were
identified. OBIA produced benthic cover maps that had higher overall accuracy, 93.78+0.85%, as compared to pixel-based methods
that had an average accuracy of only 87.30+6.11% (p-value = 0.0001, a = 0.05).
1. INTRODUCTION
1.1 Background of the study
Monitoring of coral reefs is the gathering of data and
information on ecosystems or on those who use these resources
(Hill & Wilkinson, 2004). The general process of monitoring is
identifying the population of benthic components in a reef such
as rock, rubble, algae and sand, dead or living coral
(Kenchington & Hudson, 1984 as cited in Marcos, et al., 2008).
Determining the benthic population is greatly dependent on the
scale required for assessment (Marcos, et al., 2008). For areas
of reef that need a resolution of not less than 25m? the
typically-used monitoring methods are multi-spectral satellite
imagery and aerial remote sensing (Mumby, et al, 2004).
However, such methods require ground-truthing and acquiring
such remotely-sensed images would require monetary costs.
Also, reef monitoring in many countries cover a small and
unrepresentative proportion, such that available data are
insufficient for a quantitative assessment [18]. General visual
monitoring methods are able to get information from broad to
fine scale with the advantage of using inexpensive equipment,
but these methods take a lot of time (Hill & Wilkinson, 2004).
An alternative for monitoring is the use of digital equipment,
which can greatly shorten the time in the field and reduce field
expenses, since less time is required underwater as compared to
visual methods (Hill & Wilkinson, 2004). The major drawback
of using digital equipment is that data processing, such as
digitizing, is very time consuming and equipment used are
expensive (Hill & Wilkinson, 2004). Also, accurately and
automatically mapping live benthic cover has remained
extremely difficult to produce from multi-spectral images such
as satellite images and aerial photographs, thus alternative
methods of producing these maps still need to be investigated
(Bour, et al., 1996 as stated in [18]) such as the use object-
based image analysis (OBIA). This method initially groups
pixels into objects (also called segmentation) based on certain
similarities (spectral information or external variable — such as
height) (Addink & Coillie, 2010). Rules are then developed in
order to automatically classify the image objects produced after
segmentation. With the use of OBIA, the tedious task of
digitizing and manually classifying benthic cover in the
acquired videos and photographs may be eliminated.
1.2 Objectives and significance
Objectives. This research aims to develop an improved method
of extracting benthic cover through OBIA with the use of
underwater videos and photographs with corresponding
bathymetric data. Applying the same theory used in a previous
research (Levick & Rogers, 2006) to this study, the height
component from the bathymetric data will aid in producing a
benthic cover map with better accuracy as compared to pixel-
based classification methods. The specific objectives of this
research are as follows:
e To investigate ways of georeferencing and mosaicking
snapshots of the underwater videos, as well as means of
rectifying the underwater video snapshots to the
bathymetric data, given some constraints on data
availability and quality;
e To develop the OBIA rule sets for accurately and
automatically classifying benthic cover;
To evaluate the performance of OBIA against commonly
used pixel-based image classification algorithms.
Significance. Through this automated classification. system,
fast and frequent data acquisition of benthic cover such as
living and non-living is possible to support reef studies that
