Local optimization of index assignments for multiple description coding

Index assignment problems are central to many joint source-channel coding methods and generally require a lot of computational power. We study the optimization of index assignment matrices for the multiple description coding problem. Index assignment matrices are used at the quantization step and provide a conceptually simple and fast alternative to multiply descriptive transform methods. We describe two novel local optimization algorithms using a bipartite matching procedure to locally minimize the expected mean squared error. We provide experimental results on various codebooks as well as a comparison with a previously published optimization algorithm.