Efficient Affine-invariant Fourier Descriptors for Identification of Marine Plankton

A study of population and distribution of plankton in the sea can be a good indicator of the health of the marine environment. Many digital images of marine plankton have been recorded. Image extraction and plankton identification can aid research of oceanic plankton. In this paper, we present a method to compute affine-invariant Fourier Descriptors (FDs) for marine plankton image retrieval. This method computes FDs of a shape boundary through the quasi-continuous Fourier transform. The experimental results show that the proposed FDs capture more information of the shape boundary than the the same number of traditional discrete FDs. Before calculation of FDs, each plankton image is pre-processed and the plankton shape is compacted into the boundary polygon. We have developed a set of approaches to quickly extract the exact and compact boundary polygon of an object, including methods of edge detection, boundary tracing, coordinate compensation of the boundary points and break-point detection. An affine-invariant curve normalization method then is adopted to reduce the geometrical deformations or distortions from the polygonal boundary curves possibly caused by changes of the camera angle. The experimental implementation shows that this curve normalization method is robust and can successfully eliminate transformations of translation, scaling, non-uniform scaling and shearing from two affine-transform-related curves. Lastly, the ability of the proposed FDs to identify plankton images with deformations is tested on an artificial image dataset. The experiment shows that the proposed FDs have better performance than the traditional FDs in terms of retrieval efficiency.