Distance- and neighborhood-based methods
Pattern form recognition

Here, we examine a first family of intuitive, high-performance methods based on the notion of distance between shapes. We successively study a classification method and a partitioning method, both based on this idea.

Among the multitude of RDF methods, the nearest-neighbor classification method and its immediate extension, the K-nearest-neighbor method, hold a special place. They are both natural and justified by certain theoretical results. We first describe the principle behind them, and then describe some of their properties.

Finally, we take a look at the partitioning framework and describe a common partitioning method based on the notion of distance, the K-means algorithm.