Search

We employ the small limit argument approximation (SLAA) to streamline the mathematical representation of a conventional Rician fading channel within the proposed single-input single-output (SISO) system model. The SLAA on the zeroth-ordered modified Bessel function of the first kind facilitates the transition of a newly modified Rician fading channel into the low signal-to-noise ratio (SNR) regime, where either the transmission power is low or the bandwidth is large. We reveal that, for a SISO system with an instantaneous SNR set at 20 dB and a Rician factor set at 0.05, the channel capacity amounts to 7.1 bits/s/Hz. [ABSTRACT FROM AUTHOR]

We enhanced the performance of linear prediction (LP) for bone-conducted (BC) speech signals by deriving a regularized version of the modified covariance (MC) method. The spectral estimation accuracy is discussed for the purpose of assessing the performance of LP. Inherently, BC speech has an expanded spectral dynamic range that causes ill-condition in conventional spectral estimation methods. To overcome this problem, we focus on the covariance method and its modified version, the MC method. In the field of numerical analysis, we often face the situation where an ill-conditioned case occurs in finding the solution. To deal with this situation, the regularized least squares method is utilized. Motivated by this idea, we derive the regularized MC (RMC) method in this study for BC speech analysis. The RMC method has the effect of compressing the spectral dynamic range of the input speech signal, and this effect mitigates the ill-conditioned problem of LP. Through experiments, we show that the RMC method provides a more accurate spectral estimation than conventional spectral estimation methods for BC speech where synthetic and real BC speeches are considered. The performance of the RMC method is affected by the setting of the regularization parameter. A way to determine the regularization parameter in practice is experimentally derived. The RMC method with such a setting provides the best performance of spectral estimation for BC speech. [ABSTRACT FROM AUTHOR]

In this paper, we propose the incorporation of noisy image patches and the impulse response of a low-pass filter (LPF) in a convolutional neural network (CNN) to denoise Poisson--Gaussian noise in low-dose computed tomography (LDCT) images. The approach is referred to as fast and flexible denoising CNN (FFDNet)-impulse response (FFDNet-IR) in this paper. The power spectrum sparsity LPF (SLPF) allows low-frequency components to pass through while suppressing higher frequency components by the sparsity approach of the power spectrum, and it is employed to determine the impulse response of LPF. Three well-known types of LPF, namely, Direct LPF, Gaussian LPF, and Butterworth LPF, are also considered to obtain the impulse response of LPF. In the FFDNet-IR, both the noisy image patches and the IR of the LPF are sequentially inputted into the FFDNet to eliminate the Poisson--Gaussian noise. This approach enhances the denoising performance in LDCT images compared with the conventional FFDNet in the evaluation metrics of the peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and feature similarity (FSIM). Moreover, the FFDNet-IR trained with the Poisson--Gaussian noise model demonstrates the generalization ability and effectively eliminates only Poisson or Gaussian noise. The experiments indicate that the FFDNet-IR more effectively suppresses the noise artifacts and preserves image details compared with the baseline FFDNet, as well as traditional methods such as block-matching and 3D filtering (BM3D) and nonlocal mean (NLM) for LDCT image denoising. [ABSTRACT FROM AUTHOR]

We propose a blind image quality assessment (BIQA) method of using the multitask-learningbased end-to-end convolutional neural network (CNN) approach. The architecture of the proposed method is integrated by two streams. In the first stream, multiscale image features are extracted by using the inception and pyramid pooling modules. Natural scene statistics (NSS)-based features are extracted in the second stream. The two streams are then integrated into fully connected layers to estimate the image quality score. The performance of the proposed method is validated with four public IQA databases and the obtained experimental results show the superiority of the proposed method over conventional IQA methods. [ABSTRACT FROM AUTHOR]

In this paper, we propose a blind noisy image quality estimation method of simultaneously utilizing three statistical features extracted from three different domains of the input noisy image. The statistical features used in this paper are (i) eigen-based variance by the covariance matrix of image blocks in the spatial domain, (ii) the spectral entropy of the power spectrum in the frequency domain, and the standard deviation in the wavelet domain. The extracted statistical features are fed into an extreme learning machine algorithm for mapping into perceptual quality scores. The model is trained and tested on images with six common noise distortion types commonly occurring in real-world applications: additive white Gaussian noise, additive Gaussian noise in color component, high-frequency noise, masked noise, impulse noise, and multiplicative noise. For the CSIQ, TID2008, TID2013, and KADID10k databases, the experimental results show that our method covers noise distortions wider than those of the conventional methods and achieves consistently better performance for blind noisy image quality assessment. [ABSTRACT FROM AUTHOR]

In this paper, we propose the sequential input of a noisy image patch and the impulse response of a low-pass filter (LPF) in the training of the conventional fast and flexible solution for CNN-based image denoising (FFDNet) architecture, which enhances denoising performance and edge preservation and achieves high perceptual quality. The proposed method consists of two steps. In the first step, the power spectrum sparsity is utilized to determine the impulse response of LPF and the resulting impulse response is added to the noisy image patch in a sequential form to estimate the low- and high-frequency components of the input image. In this step, the use of three different types of LPF is also considered. In the second step, the FFDNet architecture, a deep-learning-based image denoiser, is employed. The proposed method achieves satisfactory denoising performance for grayscale and color datasets on synthetic additive white Gaussian noise (AWGN) in terms of the peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), feature similarity index (FSIM), and learned perceptual image patch similarity (LPIPS) compared with the original FFDNet. The performances on realistic noise and for chest X-ray images are also investigated. [ABSTRACT FROM AUTHOR]

Speech emotion recognition has drawn extensive attention in recent years. We propose deep learning (DL)-based speech emotion recognition using synthetic bone-conducted (BC) speech. In our proposed model, air-conducted(AC) speech is transformed to BC speech using an infinite impulse response (IIR) filter. Data augmentation techniques are utilized and the parameters of convolutional neural network (CNN) models are modified to enhance the accuracy of the proposed model. Simulation results demonstrate that the proposed model outperforms the existing models in terms of recognition accuracy for BC speech. The accuracy of the proposed model is 72.50% for BC speech, whereas that of the existing model is 69.83% for AC speech. This is because BC speech can enhance low-frequency components, which is important for recognizing emotions. [ABSTRACT FROM AUTHOR]