Minimax non-redundant channel coding for vector quantization

The distortion of a vector-quantized image due to channel noise can be alleviated significantly without any redundant control bits by judicious assignment of binary codes to the codevectors. The authors consider the index assignment problem and adopt a minimax design criterion instead of the usual mean squared measure. The problem is related to the classical quadratic assignment problem and is found to be NP-hard. An effective, heuristic, polynomial-time algorithm is presented for computing approximate solutions. Simulation demonstrates that with the minimax criterion the worst-case performance is significantly improved while the average performance is only slightly degraded.<>