Adaptive compandor design using the boundary adaptation rule
Marc M. Van Hulle, Dominique Martinez · 1994
In our previous paper (1993), we introduced a novel unsupervised learning rule for scalar quantization, called the boundary adaptation rule (BAR). Adaptive quantizers were built using the maximization of information-theoretic entropy as a design criterion. In this paper, we show that BAR can also be used for designing quantizers by minimizing the mean square error distortion due to quantization. For this purpose, the adaptive histogram with equal bin counts assessed by BAR is used as a density estimator to build an optimal compandor function.>