Image Mosaicing for Neonatal Fundus Images

Abstract

Retinopathy of Prematurity (ROP) is an ocular disease observed in premature babies which, if left untreated, causes permanent blindness. Problematically, the visual indicators of ROP are not well understood and neonatal fundus images are usually of poor quality and resolution. Simplifying the methods of the detection of ROP would be highly beneficial. The various features for detection of ROP disease come from the anterior and posterior regions of the retina, which will not be available in a single image. Hence in practice, multiple images of different views are taken from the same infant and various regions are tested individually from different images. Here we propose an efficient methodology for combining the images through image mosaicing where the transformation parameters are obtained from the pre-processed neonatal fundus images. A crucial step in image mosaicing is finding robust features for feature matching, which will in turn help in obtaining the appropriate transformation parameters. In the proposed method, feature locations are taken from skeletonized images and feature descriptors from enhanced images. This approach enables us to obtain a satisfactory mosaic even while choosing a less complex six parameter-affine transformation model in contrast to the existing methodologies which require more parameters.