DARWIN Dolphin Photoidentification Software

1
DARWIN - Dolphin Photo-identification
Software Adaptations to Digital Camera
Acquisition and Increased Matching Accuracy
K. R. Debure, J. H. Stewman, S. A. Hale, Eckerd
College, St. Petersburg, FL
Automatic Outline Generation
Abstract

• The acquisition of field data with digital
  cameras requires new functionality from the
  software.
• The current version reads JPEG files and
  uncompressed TIFF format will be available in the
  next version. The automatic generation of the
  outline of the dorsal fin is one attempt to
  reduce the data entry effort. The outline is
  extracted as follows
• compute the grayscale intensity of the
  original image
• perform unsupervised thresholding based on
  histogram analysis
• use morphological processing and feature
  recognition to select and simplify
• region of interest
• use morphological processing to generate
  region outlines and select largest
• outline as dorsal fin contour

DARWIN is a computer program that automates the
photo-identification of dolphins from photographs
of dorsal fins. This program allows researchers
to query a database of previously identified
dorsal fin images with a digital image of an
unidentified dolphin's fin. The system responds
with a rank ordered list of database fin images
most closely resembling the query image. In this
way DARWIN assists researchers to prioritize
their search of database images, and potentially
reduces the time required for identification. As
digital cameras rapidly replace traditional film
cameras for the acquisition of data in the field,
adaptation of the software in response to the
changing needs of the user community is critical.
Compatibility with appropriate image formats,
efficient handling of higher resolution image
data, and automation and streamlining of data
entry can facilitate the processing of increasing
quantities of field data. The automated
generation of the outline is one example of an
effort to reduce data entry requirements. In
addition, to further increase the matching
accuracy and the realistic utility of the DARWIN
software, the most successful aspects of a
quantitative dorsal fin matching approach have
been incorporated within a broad hierarchical
approach using qualitative distinctions between
fins. This recently enhanced software (1)
facilitates subset searches of the database based
on damage category (missing tip, etc), (2)
selects more intuitive transformations for
alignment, and (3) more closely emulates the
manual photo-identification process.
Introduction
DARWIN is a computer program which aids marine
mammalogists in the identification of dolphins
and in the management of a database of
observational information. DARWIN graphically
presents the digitized dorsal fin images and
associated textual sighting information.
Outline Registration

• The database of information can be
• queried for images which are
• similar to the dorsal fin image
• of an unknown individual
• queried to limit the search within
• specified damage categories
• queried with the name of a
• specific individual
• organized and viewed by
• particular sighting information
• such as location, date, or fin
• damage category

• Direct comparison of fin outlines is problematic
  since perspective distortions can cause outlines
  extracted from multiple photographs of the same
  dolphins dorsal fin to differ significantly.
  Ideally, we would like to transform the outlines
  so that it appears the original fins were
  photographed from the same angle. The
  transformation process is as follows
• locate sets of common feature points on pairs
  of fin outlines
• compute transformation matrix which maps one
  set of points onto the other
• apply transformation to entire outline
• evaluate error between registered outlines

Figure 1 Following a query of the dorsal fin
database, DARWIN presents a list of database
images which most closely resemble the unknown
fin image.
Figure 3 Upper and center left Automatically
generated dorsal fin outlines. Upper and center
right Common feature points (the start of the
leading edge, the end of the leading edge, the
tip, the most prominent notch and the end of the
trailing edge) are identified along the outline.
Bottom left Original outlines of the database
and unknown fin are superimposed before
alignment. Bottom center Outlines are
partially registered following an affine
transform based on a triplet of feature points.
Bottom right Fully registered outlines after an
iterative alignment approach.