Fast adaptive scalar quantization for minimal mean squared error distortion in the high resolution case: extension of the boundary adaptation rule

Recently the author (1993) introduced a new unsupervised competitive learning rule for adaptive scalar quantization. The rule, called boundary adaptation rule (BAR), directly adapts the boundary points demarcating the quantization intervals and minimizes the mean absolute error distortion. In this article the author extends the BAR concept towards the mean squared error distortion minimization in a high resolution case. The performance of this extended rule is shown for stationary as well as non-stationary input probability density functions, such as speech- and image signals. The rule yields near-optimal performance.<
>