負の重みとトリムド平均を用いた線形/非線形フィルタ

Abstract

For sampled input signal, the median filter with negative weights is considered as a kind of the nonlinear filter and some application is shown. It is well known that linear filter theory based on convolution operation is used widely on the other hand. This paper tries to deal with systematically two kinds of these filters with which these character differs. The method is adopting the trimmed mean in statistics. However, since the obtained median does not necessarily come to the central position of data in the case of a median with negative weights, a special method of removing in an order from sorted data is proposed. In order to investigate these filter characteristics, the computer experiment was performed using the white Gaussian random noise input. Consequently, it turned out that the original FIR filter characteristic is realizable by this filter using trimmed mean calculation. Since this method can specify between linear and non-linear types with one parameter, it can be used as a new analyzing method in digital signal processing.