Discriminative features for bird species classification

Bird species classification has received more and more attention in the field of computer vision, for its promising applications in biology and environmental studies. Although methods derived from basic-level classification are introduced to bird species classification, most of them couldn't get a satisfied result due to the absence of discriminative features and quantization errors. In this paper, we introduce discriminative features for bird species classification based on parts of birds. We first crop and align the images, obtaining some patches specifying the parts of a bird. The patches are collected, forming some codebooks to learn the intermediate-level features using sparse coding algorithm. We then learn a model which characterize the discrimination of each part of every species of birds. Finally, the learned features combined with the model are concatenated to form the final representation for training and classification. We show the effectiveness of the discriminative features on the CUB-200- 2011 dataset. Categories and Subject Descriptors I.4.9 [Image Processing and Computer Vision]: Applications; I.5.4 [Pattern Recognition]: Applications General Terms Algorithms, Design, Experimentation.