Parasite worm egg automatic detection in microscopy stool image based on Faster R-CNN

Abstract

This paper proposed a method based on Faster R-CNN for detection of human parasite eggs in stool images. The shapes, and patterns of parasite worm in egg micro images are very diversity, therefore proposing and choosing the good model to detect them is necessary to help the doctors discover the potential disease by worm in human. To be sure for the proposal, we executed many various experiments, and retrieved dataset from two independent resources. The training set is retrieved in standard biology image library, meanwhile the evaluation image set is retrieved from real patients. The precision, recall and other values evaluated in the experiments represented the effectiveness of the method. The various experiments with the outstanding results proved the correctness of the proposal.