Binary Plankton Image Classification

In marine biology study, it is important to investigate the distribution of plankton organisms. Because of the overwhelming data size, automatic processing of the large amount of image data collected by underwater image recorders becomes inevitable. However, due to the fragmentation and the large within-class variations of binary plankton images, it is difficult to extract reliable shape features. In this paper, we propose several new shape descriptors and use a normalized multilevel dominant eigenvector estimation method to select a best feature set for binary plankton image classification. We achieve more than 91% classification accuracy in experiments on more than 3000 images