M. H. Savoji, S. Chehrehsa,
Volume 10, Issue 3 (9-2014)
Abstract
Gaussian Mixture Models (GMMs) of power spectral densities of speech and noise are used with explicit Bayesian estimations in Wiener filtering of noisy speech. No assumption is made on the nature or stationarity of the noise. No voice activity detection (VAD) or any other means is employed to estimate the input SNR. The GMM mean vectors are used to form sets of over-determined system of equations whose solutions lead to the first estimates of speech and noise power spectra. The noise source is also identified and the input SNR estimated in this first step. These first estimates are then refined using approximate but explicit MMSE and MAP estimation formulations. The refined estimates are then used in a Wiener filter to reduce noise and enhance the noisy speech. The proposed schemes show good results. Nevertheless, it is shown that the MAP explicit solution, introduced here for the first time, reduces the computation time to less than one third with a slight higher improvement in SNR and PESQ score and also less distortion in comparison to the MMSE solution.
E. Ehsaeyan,
Volume 12, Issue 2 (6-2016)
Abstract
Traditional noise removal methods like Non-Local Means create spurious boundaries inside regular zones. Visushrink removes too many coefficients and yields recovered images that are overly smoothed. In Bayesshrink method, sharp features are preserved. However, PSNR (Peak Signal-to-Noise Ratio) is considerably low. BLS-GSM generates some discontinuous information during the course of denoising and destroys the flatness of homogenous area. Wavelets are not very effective in dealing with multidimensional signals containing distributed discontinuities such as edges. This paper develops an effective shearlet-based denoising method with a strong ability to localize distributed discontinuities to overcome this limitation. The approach introduced here presents two major contributions: (a) Shearlet Transform is designed to get more directional subbands which helps to capture the anisotropic information of the image; (b) coefficients are divided into low frequency and high frequency subband. Then, the low frequency band is refined by Wiener filter and the high-pass bands are denoised via NeighShrink model. Our framework outperforms the wavelet transform denoising by %7.34 in terms of PSNR (peak signal-to-noise ratio) and %13.42 in terms of SSIM (Structural Similarity Index) for ‘Lena’ image. Our results in standard images show the good performance of this algorithm, and prove that the algorithm proposed is robust to noise.
