Clustered fractional Gabor transform

In this paper, a novel clustered fractional Gabor transform (CFrGT) is proposed by exploiting the continuity structure of the fractional Gabor spectrum. The fractional Gabor expansion is reformulated under a Bayesian framework with correlated priors. To encourage the nonzero or zero coefficients to cluster in a spatial consistent constraint, the Markov random field (MRF) model is incorporated as the prior for the support of the fractional Gabor spectrum. And then the variational Bayes expectation-maximization algorithm is used to approximate the posterior of the hidden variables and estimate the parameters of MRF model. The proposed algorithm achieves high time-frequency resolution and localization under noisy and low sampling scenarios. Both the synthetic and the real data experimental results verify the effectiveness and its superiority over the conventional fractional Gabor transform.