143 - 
ìing 
i’owl 
lies 
ìis — 
.) 
more 
etc . ). 
id 
1 
ind 
^curate 
truthing. 
ABSTRACT 
The knowledge of the size of populations of waterfowls 
in North America is necessary to properly manage this renew 
able resource. The most common methods of counting waterfowls 
during the census are by visual means from low flying aircraft 
or from aerial photographs of large bird concentrations. While 
the first method is inaccurate, the second one is time consuming 
and also presents problems of classification where a large 
number of waterfowl or a mixture of several species, age 
and sex groups occur on the photographs. To overcome these 
difficulties a research program was initiated using pattern 
recognition techniques in the census. This research covers 
three interconnected subject areas: experimental aerial 
photography, digitization of photographic images through 
micro-densitometry, and pattern recognition. 
Experimental aerial photography of greater snow geese, 
eider ducks, and of diving ducks was carried out using 
several films exposed at various scales to determine the 
best photographic media for the densitometry and pattern 
recognition. The obtained aerial photographic images were 
digitized using an automatic scanning micro-densitometer. 
Density and image shape patterns were established for the 
above species and for various sub-groups within the species. 
A FORTRAN program was developed to count black ducks and 
snow geese for one unit record of digitized image. Work 
is in progress to expand this program to accommodate larger 
records and complex patterns. 
Results indicate that normal and false color photography 
is superior to black and white in all respects. ’ The three 
different layers of the color film can be analysed separately 
by the densitometer. This yields optimum separation of 
desired groups. 
nd.