[geographomics]

Xerox multifunction printers have a feature to scan to PDF, but does so in such a way that the page image is segmented into lossy colour (DCT/JPEG) and coloured monochrome (JBIG2) depending on whether the area is detected as being photo-like or text/diagram-like respectively. These partial images are then overlaid to make the final composite page - the end result has both higher perceptual quality and considerably smaller file size then if it were encoded entirely via DCT or entirely in a lossless format such as deflate.

This is somewhat different to segmentation in video, but it does show that there are useful, fast encoding implementations of segmented image overlays.