Automatic Bird Classification from Photos
Taxonomic classification is of widespread use to biologists and non-biologists
alike, but suffers from the comparatively small supply of trained experts.
The use of automatic species identification tools is one potential solution.
Because of the large number of birdwatchers with little or no training in
taxonomy, we focused on creating an image-based automatic classifier for birds.
Bird species in photos are classified using one nearest neighbor with distances based on color and shape information in the form of color histograms and shape context features. Preliminary results show that color seems especially descriminative and performs significantly better than random at classifying bird species.
|
|
Malcolm Augat '09
|
|
