A Novel Method of Identifying Threshold for Gabor Transform Filter Based on Inter-Cluster Distance Probability

Abstract

Gabor transform suitable for time-frequency analysis and good for filtering non-stationary signals. The threshold of the Gabor transform filter is a key factor for the filter’s effectiveness. The popularly used threshold obtained by linear method is not suitable for non-stationary signals with low signal to noise ratio (SNR) because, it cannot separate the expansion coefficients of noise and useful signals. In this paper, a novel method to identify Gabor transform filter’s threshold based on initial highest inter-cluster distance probability is proposed. Simulation experiments have been carried out under several conditions. The experimental results show that the proposed threshold is highly suitable, especially when the signal’s SNR is very low and the filter output is very consistent to the real original signal and keeps no pseudo signal in zero regions.