The current minimap renders a close representation of the actual text in the file, utilizing antialiasing and sub-pixel smoothing.
In the Atom editor package minimap, the minimap text is rendered as pixel blocks. An option for this kind of a simpler algorithm could be a) pretty b) maybe even performance-enhancing on large files