Signal quantization by variation

SUMMARY Real‐valued data are quantized according to the magnitude of their fluctuations around regional mean values or around a given, real‐valued signal. Quantization does not require any parameter or threshold value except the desired number of regions. By the introduction of suitable fluctuation measures and a so‐called change point graph, the determination of a minimum quantization is transformed to the computation of the shortest path with a prescribed number of intermediate nodes. Such shortest paths are shown to be computable by dynamic programming and by a variation of the Dijkstra algorithm. Evaluations and extensions of the approach are included. Copyright © 2011 John Wiley & Sons, Ltd.