Component tree and multi-layer perceptron techniques for nanoparticle image segmentation and classification

With the development of the microscope, microscopic observations and experiments became possible; thus, fast and effective analysis of the images of cells or nanoparticles taken with high-performance microscopes has become more important than ever. The problems of particle segmentation for counting and classification by the type of particles are essential research issues that have been researched steadily so far. In this paper, we identify particle candidates for images, and we use a classifier in an attempt to classify the candidates by type. First, we build a component tree of input images in quasi-linear time and extract areas with a higher possibility of particles with their morphological features for making data set. Then, we use the data set to train multi-layer perceptron classifiers and attempt to classify the particle candidates. Experimental results showed that the particle clusters were correctly classified with high accuracy.