An Efficient Algorithm For Eyespace Classification In Go

Writing programs to play the classical Asian game of Go is considered one of the grand challenges of artificial intelligence. One of the key tactical issues in Go is whether a group of pieces can divide the area it surrounds (“eyespace”) into two separate regions, thus rendering the group immune to capture. Human players quickly learn to recognize various eyespace patterns, invariant under translation, rotation, reflection, and distortion. In this paper, we present an efficient canonical form for classifying such patterns. This representation, along with an algorithm for finding the inside of a group, is used to quickly analyze the life and death (capturability) status of all groups on the board.