Project Title: New Computerised Fluke Matching System for Humpback Whales
Chief Investigators: Dr Hendrik Kniest, A/Prof Peter Harrison and Mr Daniel Burns
The new computer-aided fluke matching system Fluke Matcher uses a unique multifaceted computer-based recognition system to overcome the overwhelming problems of manually matching photographs of humpback whale flukes in large catalogues. It provides a rapid and substantially improved method of analysing photo-identification data. The system is based on the unique characteristics of southern hemisphere humpback whale flukes, but could be adapted to suit northern hemisphere whales, as well as other marine mammal species. It significantly increases the efficiency of identifying individuals and finding resights in photo-identification catalogues of humpback whales. The system uses a wide range of criteria, based on multiple key features of humpback whale flukes that are normally utilized during manual matching methods, and additional computerised image-matching techniques, that produce a reliable matching system. Initial user input is needed to identify control points used for the transformation of the fluke image onto a common reference frame. The system then measures key features of the fluke, including parameters to describe the shape of the fluke, black and white pigment distribution in different regions of the fluke, and other distinctive features that enable identification. The system was developed and tested using a database of 117 photographs of humpback flukes with a total of 94 possible matches. Because of the broad scope of matching parameters used in the system, a number of different matching protocols can be used to rank potential matches. The optimum matching techniques resulted in 100% of the matches listed in the top 15 positions (out of 117), and 96% of the matches ranked in the top five. In this way the operator needs only to scan through a small percentage of the ranked images to find any possible match.
The first step in the process of using Fluke Matcher to analyse photo-identification data is to measure or extract the data on each photograph. As part of this process it is important to transform the imaged fluke onto a common reference system. This is accomplished by the operator marking the position of five major control points on the photograph. These points include the most easily identified points, such as the fluke tips and central V-notch and two points on the leading edge. The next stage of the program (Process Image) recalculates separate transformations for four separate districts across the fluke, but the transformed data are all defined in the one unique reference system. Angles and distances are computed that define the shape of the fluke tips and the centre V-notch area, the thickness of the black band across the trailing edge, and the general shape of the fluke. The fluke is broken up into 18 regions and the percentage of black pigmentation is calculated for each of these regions. Next the operator can measure up to 5 different types of key features that help uniquely identify the fluke. These are classified as spot, line, area, damage and image features and are one of the most important aspects in finding any match for a fluke.
A user friendly graphical interface helps the operator progress through each phase. During the search process Fluke Matcher matches each feature measured against each image in the database, giving each a score (0 to 100) for the probability that the feature properties are identical. An overall weighted Match Index (MI: 0 to 100) is then calculated for each image. The images are then ranked in order from the most likely match down to the least likely match and displayed in that order. The operator can then scan through the list to visually compare images and identify matches.
