VARIABLE METRIC FOR BINARY VECTOR QUANTIZATION (WP-P8)

Author(s) : Ismo Kärkkäinen (Department of Computer Science, University of Joensuu, Finland) Pasi Fränti (Department of Computer Science, University of Joensuu, Finland)

Abstract : We present a new method for performing vector quantization of binary vectors. The proposed method varies the distance metric and updates the centroids in an optimal manner regarding the current metric. The corresponding centroids change from “soft”, allowing variables of codevectors to have values between zero and one, to hard, pure binary codevectors.

Menu