Multiple Description Image Coding Using Regions of Interest

Abstract

A multiple description (MD) coding scheme encodes an image such that each description alone yields an acceptable reconstruction quality, and the descriptions together render increasingly better approximations of the image. Traditionally, each description contributes equally to the reconstruction quality throughout the image. This is inefficient from a coding perspective since all descriptions largely represent similar source information. We suggest a novel region of interest (ROI) based MD coding scheme with two descriptions and two ROIs, such that each ROI is coded into a separate description. The impact this scheme has on visual reconstruction quality and coding efficiency is investigated. We argue that ROI based MD coding is beneficial at low bit rates since it (i) introduces little redundancy between the descriptions with practically no loss in quality at the central decoder, and (ii) still produces useful side descriptions. Extensions to video coding are discussed in the concluding section.