Automatic segmentation of Chaetoceros microscopic images via pixel-wise classification

Chaetoceros is one of the largest marine genus of phytoplankton and plays an important role in ocean ecosystem and global climate. Automatic segmentation of Chaetoceros is a challenging task due to the species' biomorphic characteristics. In this paper, we present a novel method using pixel-wise classification by Support Vector Machine (SVM) [1] combined with Grayscale Surface Direction Angle Model (GSDAM) [2] for the segmentation of Chaetoceros microscope images. Firstly, the pixel-level features are acquired by the five maps of GSDAM, which are used as the input of SVM classifier. Secondly, the SVM classifier is trained by the pixel samples automatically selected from connected region pre-segmentation result. Then, the trained SVM classifier is applied to classify all the image pixels into two classes (Object and Background). Finally some post-processing are done to generate the final segmentation map. Our method not only takes advantage of biomorphic characteristics information of Chaetoceros, but also the ability of SVM classification. Experimental results show that our method achieves effective segmentation performance, yielding high setae information reserving.