Contour Extraction of Drosophila Embryos Based on Conditional Generative Adversarial Nets

Abstract

High-quality Drosophila embryo images can provide reliable data sources for the research of gene expression and gene interaction. Based on the Drosophila embryo images, a FEMine system is constructed to assist geneticists in quickly mining information for analysis. The extraction of interested Drosophila embryo is an important pretreatment in the FEMine systems. In this paper, taking contour extraction task as image generation task, we proposed a conditional generative adversarial network to generate contour maps of the same size as input images. Based on the Drosophila embryo manual dataset DEDS(Drosophila embryo Dataset)1, for each of the ground-truth, we turn the regions of the interested Drosophila embryo into contours of that. The experiments on DEDS demonstrate that our framework can efficiently extract the contours of the interested Drosophila embryos.