Pairwise Nonparametric Discriminant Analysis for Binary Plankton Image Recognition

Plankton image classification is an important, yet challenging, problem in marine biology. This challenge can be attributed to: large within-class variations; large between-class similarity; and large noise. To mitigate these problems, we propose a novel subspace classification framework, called pairwise nonparametric discriminant analysis for binary plankton image recognition. In this framework, first we decompose the multiclass recognition into a combination of pairwise binary classes, then train an appropriate classifier for each class pair using the nonparametric discriminant analysis technique (a newly developed subspace analysis technique) to effectively remove unwanted information (such as the within-class variations and the noise) and extract discriminant information (such as the boundary structural information), and, finally, combine all the pairwise classifiers using an efficient fusion rule for real-time classification. Extensive experiments are conducted on a large data set to show the improvement obtained by our new approach over the state-of-the-art ones.