Automatic non-parametric capsid segmentation using wavelets transform and graph

Every year, one million people dies from Hepatitis B virus. As for other viruses, its genetic material is enclosed by a capsid whose segmentation and classification is essential. In this paper, we present a novel capsid segmentation technique which is a combination of a “à trous” wavelet process (for background filtering) and a graph-based structure (for segmentation and classification). Capsids were acquired in transmission electron microscopy (TEM) as a set of 9 series of 40 images each. Our technique achieved as much as 80% of capsid detection and classification for 8 of the series, reaching more than 90% for 5 of them.