would it be possible to have layers of emphasis levels on the annotations, e.g. "primary annotations" stand out (as now) with an attention-grabbing outline, whilst "secondary annotations" only have a tooltip visible as you glide the cursor around. When overlapping, the 'primary annotations' would always take precedence.
This would allow people to saturate images with links, allowing maximum discovery without adding visual clutter. Note the possibility of wikipedia users discovering images via 'what links here'.if you link *everything imaginable*, others may discover connections that no individual initially spotted.
the images could function both as a 'visual dictionary', or as translation hints (e.g. imagine if many phrases found in wikipedia have a series of images available to clarify what they are)