Automatic standardisation of a zebrafish embryo image database

Recently, there has been an increasing interest to propose computational approaches based on image processing to automate the comparison of spatial gene expression patterns contained in transgenic embryo images. The first step is typically to classify the images between lateral and dorsal/ventral views, and then to align them along the anterior and posterior ends. While some methodologies have been recently proposed for standardizing images in Drosophila database, the case of zebrafish embryos have never been tackled. In this paper, we propose a standardisation methodology for building a zebrafish embryos image database. Details of our approach and the results of using this approach on a pilot dataset are presented. A quality test first allows rejecting images presenting artefacts and which cannot be processed. The retained images are then classified, achieving a success rate of 93%.