Efficiently compressing an image of an object from a segmentation mask

Question

Are there any suggestions on efficient ways to compress an image, out of which you only want to save information of 1 segmented class and disregard the other.

Conside