Quantifying spatial relationships from whole retinal images