Multiple description image coder using correlating transforms

We propose a multiple description image coder which uses a linear transform to generate different correlated descriptions. This coding technique is useful when transform coefficients have to be transmitted over lossy channels, such as packet networks, since statistical dependencies among descriptions help the receiver to better estimate lost coefficients. The correlating transform is determined from a gradient search and assumes knowledge of the channel loss statistics as described in [1]. Simulation results show a qualitatively and quantitatively considerable performance increase over a standard DCT based image encoder.