Protozoa Identification using 3D Geometric Multiple Color Channel Local Feature

Protozoa identification plays an important role in parasitology in order to give right analyses and treatments. Previous identification methods analyze protozoa using morphology on microscopic images. In this work, we focus on a set of specific species which cause disease to human such asGiardia lambia, Iodamoeba butchilii, Toxoplasma gondi, Cyclospora cayetanensis, Balantidium coli, Sarcocystis, Cystoisospora belli and Acanthamoeba. They share the similar rounded shapes that make it difficult to identify them. We propose 3D Geometric Multiple Color Channel Local Feature to characterize these species of protozoa. Our method groups key points of the local feature with respect to 2D geometric in planar images and scale-level from multiple color channels of input images to produce feature vectors for identification task. Support Vector Machine is used as the classifier to identify the species of protozoa. Experiments on our dataset show the potential of our proposed method with a good performance of 96% of accuracy and outperformed previous methods.