An image analysis example using learning agents

When fully automatic image analysis is not feasible, an interactive, semiautonomous system may suffice. We present an application in electron microscopy in which the user gives the computer varying degrees of autonomy to detect gold-stained proteins in images of the surface of a cell. Initially, the system operates in a manual mode: the user clicks the mouse on the image to designate the coordinates of a protein molecule. In the most autonomous mode, a single mouse click inside a cluster of particles spawns an agent which "nests" in a molecule and "breeds": producing copies of itself each of which searches for a molecule. The user always has the option to override the autonomous labeling, so classification accuracy is assured.