scholarly journals Adaptive Wiener Filter and Natural Noise to Eliminate Adversarial Perturbation

Electronics ◽  
2020 ◽  
Vol 9 (10) ◽  
pp. 1634
Author(s):  
Fei Wu ◽  
Wenxue Yang ◽  
Limin Xiao ◽  
Jinbin Zhu
Keyword(s):  
Speech Processing ◽  
Wiener Filter ◽  
Window Size ◽  
Saliency Map ◽  
Wiener Filtering ◽  
Local Variance ◽  
Optimal Estimator ◽  
Natural Noise ◽  
Multiple Alternative ◽  
Adversarial Examples

Deep neural network has been widely used in pattern recognition and speech processing, but its vulnerability to adversarial attacks also proverbially demonstrated. These attacks perform unstructured pixel-wise perturbation to fool the classifier, which does not affect the human visual system. The role of adversarial examples in the information security field has received increased attention across a number of disciplines in recent years. An alternative approach is “like cures like”. In this paper, we propose to utilize common noise and adaptive wiener filtering to mitigate the perturbation. Our method includes two operations: noise addition, which adds natural noise to input adversarial examples, and adaptive wiener filtering, which denoising the images in the previous step. Based on the study of the distribution of attacks, adding natural noise has an impact on adversarial examples to a certain extent and then they can be removed through adaptive wiener filter, which is an optimal estimator for the local variance of the image. The proposed improved adaptive wiener filter can automatically select the optimal window size between the given multiple alternative windows based on the features of different images. Based on lots of experiments, the result demonstrates that the proposed method is capable of defending against adversarial attacks, such as FGSM (Fast Gradient Sign Method), C&W, Deepfool, and JSMA (Jacobian-based Saliency Map Attack). By compared experiments, our method outperforms or is comparable to state-of-the-art methods.

Regression-Based Noise Modeling for Speech Signal Processing

2021 ◽  
pp. 2150022
Author(s):  
Caio Cesar Enside de Abreu ◽  
Marco Aparecido Queiroz Duarte ◽  
Bruno Rodrigues de Oliveira ◽  
Jozue Vieira Filho ◽  
Francisco Villarreal
Keyword(s):  
Speech Enhancement ◽  
Speech Processing ◽  
Acoustic Analysis ◽  
Voice Quality ◽  
Wiener Filter ◽  
Processing System ◽  
Speech Quality ◽  
Speech Signals ◽  
Speech Signal Processing ◽  
Acoustic Environment

Speech processing systems are very important in different applications involving speech and voice quality such as automatic speech recognition, forensic phonetics and speech enhancement, among others. In most of them, the acoustic environmental noise is added to the original signal, decreasing the signal-to-noise ratio (SNR) and the speech quality by consequence. Therefore, estimating noise is one of the most important steps in speech processing whether to reduce it before processing or to design robust algorithms. In this paper, a new approach to estimate noise from speech signals is presented and its effectiveness is tested in the speech enhancement context. For this purpose, partial least squares (PLS) regression is used to model the acoustic environment (AE) and a Wiener filter based on a priori SNR estimation is implemented to evaluate the proposed approach. Six noise types are used to create seven acoustically modeled noises. The basic idea is to consider the AE model to identify the noise type and estimate its power to be used in a speech processing system. Speech signals processed using the proposed method and classical noise estimators are evaluated through objective measures. Results show that the proposed method produces better speech quality than state-of-the-art noise estimators, enabling it to be used in real-time applications in the field of robotic, telecommunications and acoustic analysis.

Smoothing imaging condition for shot-profile migration

Geophysics ◽  
2007 ◽  
Vol 72 (3) ◽  
pp. S149-S154 ◽  
Author(s):  
Antoine Guitton ◽  
Alejandro Valenciano ◽  
Dimitri Bevc ◽  
Jon Claerbout
Keyword(s):  
Time Domain ◽  
Signal To Noise Ratio ◽  
Wiener Filter ◽  
Window Size ◽  
Trial And Error ◽  
Data Set ◽  
Smoothing Operator ◽  
Imaging Condition ◽  
Damping Parameter ◽  
The Time Domain

Amplitudes in shot-profile migration can be improved if the imaging condition incorporates a division (deconvolution in the time domain) of the upgoing wavefield by the downgoing wavefield. This division can be enhanced by introducing an optimal Wiener filter which assumes that the noise present in the data has a white spectrum. This assumption requires a damping parameter, related to the signal-to-noise ratio, often chosen by trial and error. In practice, the damping parameter replaces the small values of the spectrum of the downgoing wavefield and avoids division by zero. The migration results can be quite sensitive to the damping parameter, and in most applications, the upgoing and downgoing wavefields are simply multiplied. Alternatively, the division can be made stable by filling the small values of thespectrum with an average of the neighboring points. This averaging is obtained by running a smoothing operator on the spectrum of the downgoing wavefield. This operation called the smoothing imaging condition. Our results show that where the spectrum of the downgoing wavefield is high, the imaging condition with damping and smoothing yields similar results, thus correcting for illumination effects. Where the spectrum is low, the smoothing imaging condition tends to be more robust to the noise level present in the data, thus giving better images than the imaging condition with damping. In addition, our experiments indicate that the parameterization of the smoothing imaging condition, i.e., choice of window size for the smoothing operator, is easy and repeatable from one data set to another, making it a valuable addition to our imaging toolbox.

Synchrosqueezing Transform Based Powerline Interference Reduction in ECG Recording

2020 ◽  
Vol 10 (10) ◽  
pp. 2259-2273
Author(s):  
M. Suresh Kumar ◽  
G. Krishnamoorthy ◽  
D. Vaithiyanathan
Keyword(s):  
Operation Mode ◽  
Wiener Filter ◽  
Wiener Filtering ◽  
Ecg Signal ◽  
Intrinsic Mode Functions ◽  
Ecg Signals ◽  
Mode Decomposition ◽  
Synchrosqueezing Transform ◽  
Signal Mode ◽  
Powerline Interference

This paper presents an adaptive ECG enhancement procedure based on Synchrosqueezing Transform (SST) to eliminate Powerline interference (PLI) from ECG signal. This work also incorporates the principles of modified discrete cosine transform (MDCT) and wiener filter. PLI is a major source of artifacts in the ECG signal which can affect its interpretation. Separating PLI from ECG signal poses a great challenge in the ECG analysis. The existing PLI removal techniques suffer from two major drawbacks such as Mode Mixing, inability to deal with non-stationary characteristics of signal. In this paper, we propose SST based wiener filtering approaches which can overcome the limitation of existing PLI suppression techniques. The proposed approaches undergo three stages of operation: mode decomposition, mode determination and peak restoration to filter out PLI from ECG recording. The mode decomposition uses SST to decompose the corrupted ECG signal into a sum of well separated intrinsic mode functions (IMFs). The objective is to filter out PLI from these IMFs. To do so, mode determination step which is based on Kurtosis and Crest factor is applied to categorize decomposition result into groups such as signal mode and noisy mode. Direct subtraction of the noisy mode from the corrupted ECG observation results in ECG signal with reduced peak since noise mode carries part of signal components in addition to interference. Hence, to restore the peak, wiener filter is applied on noisy mode to estimate actual PLI component. Finally, Noise free ECG signal is reconstructed by subtracting estimated PLI from the corrupted ECG signal. This paper discusses four possible PLI suppression methods which are derived by combining SST domain with wiener filter in various ways. Simulations are carried out to test the effectiveness of proposed methods. It is evident from the simulation results that the proposed methods can remove PLI of 50 Hz and its harmonics. The proposed techniques effectively removed PLI in both real and artificial ECG signals and to test its performance they are compared with state of the art methods. The SST based filtering methods outperformed other methods under the condition of PLI frequency variations. The experimental results also suggest that the SST based wiener filtering with modified reference approach offers better PLI suppression than all other methods.

scholarly journals Deconvolution of Defocused Image with Multivariate Local Polynomial Regression and Iterative Wiener Filtering in DWT Domain

2010 ◽  
Vol 2010 ◽  
pp. 1-14 ◽  
Author(s):  
Liyun Su ◽  
Fenglan Li
Keyword(s):  
Polynomial Regression ◽  
Wiener Filter ◽  
Simulated Data ◽  
Wiener Filtering ◽  
Image Deconvolution ◽  
Local Polynomial Regression ◽  
Local Polynomial ◽  
Multivariate Local Polynomial Regression ◽  
Restoration Method ◽  
Defocused Image

A novel semiblind defocused image deconvolution technique is proposed, which is based on multivariate local polynomial regression (MLPR) and iterative Wiener filtering (IWF). In this technique, firstly a multivariate local polynomial regression model is trained in wavelet domain to estimate defocus parameter. After obtaining the point spread function (PSF) parameter, iterative wiener filter is adopted to complete the restoration. We experimentally illustrate its performance on simulated data and real blurred image. Results show that the proposed PSF parameter estimation technique and the image restoration method are effective.

Generating facial expression adversarial examples based on saliency map

2021 ◽  
pp. 104318
Author(s):  
Yudao Sun ◽  
Juan Yin ◽  
Chunhua Wu ◽  
KangFeng Zheng ◽  
XinXin Niu
Keyword(s):  
Facial Expression ◽  
Saliency Map ◽  
Adversarial Examples

Research of Denoising Technology about Wavelet Analysis with Wiener Filter

2013 ◽  
Vol 333-335 ◽  
pp. 597-600
Author(s):  
Yao Bin Hu ◽  
Liang Bin Hu ◽  
Qiang Cheng
Keyword(s):  
Signal Processing ◽  
Wavelet Analysis ◽  
White Noise ◽  
Frequency Spectrum ◽  
Wiener Filter ◽  
Power Line ◽  
Wiener Filtering ◽  
Power Line Communication ◽  
Denoising Method

Interfering noise of power line is one of the important factors which affects the quality of power line communication (PLC). Its frequency spectrum has the character of the 1/f process and the great autocorrelation. The wavelet analysis is an important signal-processing tool. Selecting suitable wavelet analysis can turn non-white noise to white noise, followed by wiener filtering, we can achieve the purpose of denoising. This paper introduces a denoising method of combining wavelet analysis with wiener filtering. Experiment proves this method has a strong feasibility and practical value.

Research on the Algorithm of Adaptive Loop Filter Based on Flexible Quadtree

2013 ◽  
Vol 423-426 ◽  
pp. 2491-2495
Author(s):  
Gang Wang ◽  
He Xin Chen ◽  
Mian Shu Chen
Keyword(s):  
Wiener Filter ◽  
Wiener Filtering ◽  
Reconstructed Image ◽  
Optimal Size ◽  
Original Image ◽  
Deblocking Filter ◽  
Loop Filter ◽  
Coding Quality ◽  
Quad Tree ◽  
Adaptive Loop Filter

The algorithm of adaptive loop filter based on flexible quad tree (Q-ALF) is put forward targeting the problem that deblocking loop filter used by H.264/AVC cant further improve video coding quality. This algorithm, through the original image and the image after deblocking filter, works out wiener filter and selects a wiener filter with optimal size by taking the SAD of the image after Wiener filtering and the original image as RDO cost and based on RDO model, then uses this filter to filtrate the area that needs filtration. Simulation result shows that Q-ALF algorithm can improve the PSNR of the reconstructed image and decrease bit rate; meanwhile, Q-ALF algorithm can better eliminate blocking effect, improving the subjective performance of the reconstructed image.

Characterizing and correcting for the dynamic response of a bag-in-box system

1984 ◽  
Vol 56 (1) ◽  
pp. 254-258
Author(s):  
J. H. Bates ◽  
G. K. Prisk ◽  
T. E. Tanner ◽  
A. E. McKinnon
Keyword(s):  
Impulse Response ◽  
Flow Measurement ◽  
Gas Flow ◽  
Wiener Filter ◽  
Exponential Model ◽  
Wiener Filtering ◽  
Computational Techniques ◽  
Gas Flows ◽  
Order Model ◽  
Model Method

A bag-in-box system (BBS) whose volume is monitored by a mechanical spirometer tends to have a slow response if the volume of the box is large, and this may significantly affect its measurement of gas flow. We describe a device for creating reproducible gas flows with which the impulse response of a BBS may be conveniently determined. Two computational techniques for correcting a BBS flow measurement for the effects of the impulse response were investigated: 1) an exponential model method that assumes a second-order model of the BBS dynamics and 2) a Fourier transform-based method of deconvolution known as Wiener filtering. Both correction methods produced a significant increase in the accuracy of BBS flow estimations, with the Wiener filter giving superior results.

scholarly journals A New Neutrosophic Approach of Wiener Filtering for MRI Denoising

2013 ◽  
Vol 13 (4) ◽  
pp. 177-186 ◽  
Author(s):  
J. Mohan ◽  
V. Krishnaveni ◽  
Yanhui Guo
Keyword(s):  
Magnetic Resonance ◽  
Wiener Filter ◽  
Magnetic Resonance Images ◽  
Wiener Filtering ◽  
Neutrosophic Set ◽  
Rician Noise ◽  
Total Variation Minimization ◽  
Image Domain ◽  
Anisotropic Diffusion Filter ◽  
Local Means

In this paper, a new filtering method based on neutrosophic set (NS) approach of wiener filter is presented to remove Rician noise from magnetic resonance image. A neutrosophic set, a part of neutrosophy theory, studies the origin, nature and scope of neutralities, as well as their interactions with different ideational spectra. Now, we apply the neutrosophic set into image domain and define some concepts and operators for image denoising. The image is transformed into NS domain, described using three membership sets: True (T), Indeterminacy (I) and False (F). The entropy of the neutrosophic set is defined and employed to measure the indeterminacy. The ω-wiener filtering operation is used on T and F to decrease the set indeterminacy and remove noise. The experiments have conducted on simulated Magnetic Resonance images (MRI) from Brainweb database and clinical MR images corrupted by Rician noise. The results show that the NS wiener filter produces better denoising results in terms of visual perception, qualitative and quantitative measures compared with other denoising methods, such as classical wiener filter, the anisotropic diffusion filter, the total variation minimization scheme and non local means filter.

scholarly journals Improvement of binarization performance using local otsu thresholding

2019 ◽  
Vol 9 (1) ◽  
pp. 264
Author(s):  
Khairun Saddami ◽  
Khairul Munadi ◽  
Yuwaldi Away ◽  
Fitri Arnia
Keyword(s):  
Character Recognition ◽  
Wiener Filter ◽  
Transformation Method ◽  
Wiener Filtering ◽  
Median Filtering ◽  
Important Process ◽  
Connected Component ◽  
Image Subtraction ◽  
Otsu Thresholding ◽  
Processing Steps

<p><span>Ancient document usually contains multiple noises such as uneven-background, show-through, water-spilling, spots, and blur text. The noise will affect the binarization process. Binarization is an extremely important process in image processing, especially for character recognition. This paper presents an improvement to Nina binarization technique. Improvements were achieved by reducing processing steps and replacing median filtering by Wiener filtering. First, the document background was approximated by using Wiener filter, and then image subtraction was applied. Furthermore, the manuscript contrast was adjusted by mapping intensity of image value using intensity transformation method. Next, the local Otsu thresholding was applied. For removing spotting noise, we applied labeled connected component. The proposed method had been testing on H-DIBCO 2014 and degraded Jawi handwritten ancient documents. It performed better regarding recall and precision values, as compared to Otsu, Niblack, Sauvola, Lu, Su, and Nina, especially in the documents with show-through, water-spilling and combination noises.</span></p>

Sign in / Sign up

Export Citation Format

Share Document