Effect of speech-intrinsic variations on human and automatic recognition of spoken phonemes

The aim of this study is to quantify the gap between the recognition performance of human listeners and an automatic speech recognition (ASR) system with special focus on intrinsic variations of speech, such as speaking rate and effort, altered pitch, and the presence of dialect and accent. Second, it is investigated if the most common ASR features contain all information required to recognize speech in noisy environments by using resynthesized ASR features in listening experiments. For the phoneme recognition task, the ASR system achieved the human performance level only when the signal-to-noise ratio (SNR) was increased by 15 dB, which is an estimate for the human-machine gap in terms of the SNR. The major part of this gap is attributed to the feature extraction stage, since human listeners achieve comparable recognition scores when the SNR difference between unaltered and resynthesized utterances is 10 dB. Intrinsic variabilities result in strong increases of error rates, both in human speech recognition (HSR) and ASR (with a relative increase of up to 120%). An analysis of phoneme duration and recognition rates indicates that human listeners are better able to identify temporal cues than the machine at low SNRs, which suggests incorporating information about the temporal dynamics of speech into ASR systems.     

小目标识别的小波阈值去噪方法

为改善小目标识别的滤噪效果并提高其信噪比,构造了新的阈值函数并采用局部方差估计法来计算阈值对小目标进行去噪处理。对小波分解层次中各高频子带选取不同的阈值,其中大于阈值的小波系数采用改进的双曲线函数作为阈值函数,小于阈值的小波系数采用指数函数与对数函数相互组合的方式作为阈值函数。对采用的阈值函数进行了理论推导,并与软、硬阈值法进行了实验对比。计算机仿真结果表明：经本文阈值法处理后,信噪比相对于含噪图像提高了70.8%,而软、硬阈值法分别提高了49.8%和59.7%。光学实验进一步证实：该方法能更有效地提高信噪比,增强联合变换相关器对于小目标的识别能力。     

Higher-order cross-correlation-based Doppler optical coherence tomography

A method based on higher-order cross-correlation is proposed to fetch the Doppler information on flow velocity within areas under low signal-to-noise ratio (SNR) by spectral domain optical coherence tomography. The proposed method is theoretically developed and validated by measurement of a moving mirror with known velocities. Standard deviations of flow velocities of the mirror under different SNRs are determined by the proposed method and compared with those by the modified phase-resolved method. Measurement of flowing particles within a glass capillary is also conducted, and Doppler flow velocity maps of the glass capillary are reconstructed by both methods. All experimental results demonstrate that the proposed method can significantly suppress noise, thus rendering it suitable for flow measurement under low SNR cases.     

Comparison of simultaneous signals obtained from a dual-field-of-view lidar and its application to noise reduction based on empirical mode decomposition

Although the empirical mode decomposition (EMD) method is an effective tool for noise reduction in lidar signals, evaluating the effectiveness of the denoising method is difficult. A dual-field-of-view lidar for observing atmospheric aerosols is described. The backscattering signals obtained from two channels have different signal-to-noise ratios (SNRs). The performance of noise reduction can be investigated by comparing the high SNR signal and the denoised low SNR signal without a simulation experiment. With this approach, the signal and noise are extracted to one intrinsic mode function (IMF) by the EMD- based denoising; thus, the threshold method is applied to the IMFs. Experimental results show that the improved threshold method can effectively perform noise reduction while preserving useful sudden-change information.     

基于梯度一致性约束的多光谱/全色影像最大后验融合方法

多光谱/全色影像融合可以得到高空间分辨率的多光谱影像,在影像解译和分类等方面具有十分重要的意义。提出一种基于梯度一致性约束的遥感影像融合方法。该方法在最大后验概率框架下,通过梯度一致性约束建立理想高空间分辨率多光谱影像和全色影像之间的关系,并结合多光谱影像观测模型和Huber-Markov影像先验,构建融合目标函数,最后采用梯度下降法求解得到融合影像。本文方法在目标函数中引入了梯度一致性约束,克服了现有的同类方法受限于波段数量的缺陷,同时在求解中自适应确定每个波段的迭代步长,充分顾及了各波段的光谱特性,从而既确保了融合影像的光谱信息保真度,也提高了融合影像的空间信息融入度。通过IKONOS和WorldView-2影像对该方法进行了验证,并和GS,AIHS和AMBF等融合方法从定性和定量两方面进行了比较分析。实验结果表明,相比于其他方法,该方法可以在更好保持光谱信息的同时增强影像的空间分辨率,具有更广泛的适用范围和更佳的融合效果。     

旋转不变的非局域均值算法在磁共振图像去噪中的应用

磁共振成像（MRI）实验时常采用多次扫描累加平均提高图像信噪比（SNR）,但当扫描过程中运动引起图像变形时,简单地累加平均就无法奏效.为此,本研究组曾提出一种匹配加权平均方法（MWA）提高图像的信噪比.在此基础上,该文提出一种旋转不变的非局域均值算法（RINLM）,即选取圆形邻域区域并将其划分为一系列以中心像素为圆心的等面积圆环,再计算模式的相似性.RINLM算法可以更好地利用图像中旋转的冗余信息、找到更多的相似结构,提高算法的去噪性能.我们把该方法应用于低信噪比图像序列的平均和去噪中,可以更好地处理旋转的局部运动.与非局域均值算法（NLM）相比,RINLM算法可以进一步提高图像的信噪比;与MWA方法相比,其与RINLM算法的结合可以进一步提高磁共振图像序列信噪比,更好的保持图像边缘信息.     

Noise analysis of grating-based x-ray differential phase-contrast imaging with angular signal radiography

X-ray phase-contrast imaging is one of the novel techniques,and has potential to enhance image quality and provide the details of inner structures nondestructively.In this work,we investigate quantitatively signal-to-noise ratio(SNR) of grating-based x-ray phase contrast imaging(GBPCI) system by employing angular signal radiography(ASR).Moreover,photon statistics and mechanical error that is a major source of noise are investigated in detail.Results show the dependence of SNR on the system parameters and the effects on the extracted absorption,refraction and scattering images.Our conclusions can be used to optimize the system design for upcoming practical applications in the areas such as material science and biomedical imaging.     

Toward automatic evaluation of defect detectability in infrared images of composites and honeycomb structures

Non-destructive testing (NDT) refers to inspection methods employed to assess a material specimen without impairing its future usefulness. An important type of these methods is infrared (IR) for NDT (IRNDT), which employs the heat emitted by bodies/objects to rapidly and noninvasively inspect wide surfaces and to find specific defects such as delaminations, cracks, voids, and discontinuities in materials. Current advancements in sensor technology for IRNDT generate great amounts of image sequences. These data require further processing to determine the integrity of objects. Processing techniques for IRNDT data implicitly looks for defect visibility enhancement. Commonly, IRNDT community employs signal to noise ratio (SNR) to measure defect visibility. Nonetheless, current applications of SNR are local, thereby overseeing spatial information, and depend on a-priori knowledge of defect’s location. In this paper, we present a general framework to assess defect detectability based on SNR maps derived from processed IR images. The joint use of image segmentation procedures along with algorithms for filling regions of interest (ROI) estimates a reference background to compute SNR maps. Our main contributions are: (i) a method to compute SNR maps that takes into account spatial variation and are independent of a-priori knowledge of defect location in the sample, (ii) spatial background analysis in processed images, and (iii) semi-automatic calculation of segmentation algorithm parameters. We test our approach in carbon fiber and honeycomb samples with complex geometries and defects with different sizes and depths.     

Investigations of range accuracy for long-range three-dimensional active imaging

Distance resolutions and noises are analyzed experimentally for long-range three-dimensional (3D) active imaging systems that have signal-to-noise ratios (SNRs) more optimal than 30:1. Findings indicate that the photon shot noise primarily determines the SNR. However, the active imaging method, which has a relatively low SNR, generates a relatively high distance resolution. To explain this phenomenon, a theory in which the distance resolution of 3D active imaging systems is determined by both the photon shot noise and the subinterval width is developed. Theoretical and experimental results differ by less than 4%.     

基于局部位移校正的磁共振图像相干平均

多次扫描相干平均是提高磁共振图像信噪比的常用方法,但如果在多次扫描过程中病人发生自主或不自主的运动,使得图像中的组织发生位移,简单相干平均图像会导致图像模糊.本文受非局域均值算法的启发,提出了一种基于局部位移校正的相干平均方法.该算法通过比较多次采集的图像中组织结构的局部相似性,找出图像间的局部位移,利用该信息修正位移后进行加权平均,从而达到提高图像信噪比的目的.我们用模型及真实的肝脏弥散数据进行了实验.实验结果表明,对于不同次采样间存在运动的磁共振图像,该算法可有效地提高信噪比并保持结构边缘;其结果优于简单的相干平均,去噪效果也优于经典的非局域均值算法.     

基于非均匀周期采样的傅里叶望远镜时域信号采集方法

为了在低信噪比条件下清晰重构深空暗弱目标,提出了一种基于非均匀周期采样（NUPS）的傅里叶望远镜（FT）时域信号采集方法。对提出的方法进行了模拟实验并与均匀采样方法重构的图像进行了对比。基于NUPS方法,用1 MHz和5 MHz的采样频率分别采集100个点,对两个序列信号分别进行快速傅里叶变换,并对关心频率信息进行平均;传统的均匀采样方法则分别用1 MHz和5 MHz的采样频率采集200个点,再进行解调平均。对比结果显示：当信噪比（SNR）为50时,本文重构图像与衍射极限图像的斯托里尔比（Strehl）相比原方法提升了0.03,SNR为20时,Strehl比为0.531 1,较均匀采样提高了0.223 3。实验结果表明：NUPS方法在低信噪比条件下成像质量较高,可降低对激光功率的要求,为FT工程系统的实施奠定了技术基础。     

Chaotic modulations and performance analysis for digital underwater acoustic communications

Two modulation schemes, M-ary phase shift keying (MPSK) and M-ary frequency shift keying (MFSK), are commonly used for coherent and incoherent digital communications, respectively. Despite wide applications in underwater acoustic (UWA) communications, they are not suitable for confidential applications for their well-known generating processes and signal features. In this paper, two corresponding chaotic modulation methods are proposed to improve their security, namely chaotic MPSK (CMPSK) and chaotic MFSK (CMFSK). By application of chaotic sequences into the modulation procedures, they can prevent the unauthorized receivers from extracting information from the intercepted signals even with high SNR. The confidential performance of chaotic modulations is evaluated by a designed automatic modulation classification (AMC) system. Simulation results indicate a success identification rate of more than 90% for the MPSK and MFSK signal at the SNR from −10 dB to 40 dB, but only an identification rate of nearly zero for the CMPSK and CMFSK signal. Therefore, chaotic modulations have lower available probability and can achieve higher confidential performance. Also, an experiment was conducted to verify the performance of chaotic modulations in actual UWA communications. The experimental results show that chaotic modulations can achieve similar bit error ratio (BER) compared with conventional digital modulations, which verifies potential applications of chaotic modulations in confidential UWA communications.     

Evidence specifically favoring the equalization-cancellation theory of binaural unmasking

Three experiments investigated the roles of interaural correlation (rho) and of the monaural power spectrum in the detection and discrimination of narrow-band-noise signals (462-539 Hz) in broadband maskers (0-3 kHz). The power and rho of the target band were independently controlled, while the flanking noise was fixed and diotic. Experiments 1 and 2 involved rho and power values that would be produced by specific values of signal-to-noise ratio (SNR) in the NoSpi binaural configuration. Listeners were required to discriminate different SNRs via a 2I-FC loudness-discrimination task. At low reference SNRs, changes in rho fully accounted for listeners' performance, but as reference SNR increased, additional energy in the target band played an increasing role. Experiment 2 showed that at these higher SNRs the combination of information from the power spectrum and rho was superadditive and could not be explained by simple signal-detection models. The equalization-cancellation (EC) theory would explain these data using the output from interaural cancellation, Y, rather than rho. Experiment 3 attempted to foil binaural processing, by fixing either rho or Y across intervals. Consistent with EC theory, when Y was fixed, the contribution of the binaural system appeared negligible, while fixing rho did not have this effect.     

Blind parameter estimation of pseudo-random binary code-linear frequency modulation signal based on Duffing oscillator at low SNR

Conventional parameter estimation methods for pseudo-random binary code-linear frequency modulation(PRBCLFM) signals require prior knowledge, are computationally complex, and exhibit poor performance at low signal-to-noise ratios(SNRs). To overcome these problems, a blind parameter estimation method based on a Duffing oscillator array is proposed. A new relationship formula among the state of the Duffing oscillator, the pseudo-random sequence of the PRBC-LFM signal, and the frequency difference between the PRBC-LFM signal and the periodic driving force signal of the Duffing oscillator is derived, providing the theoretical basis for blind parameter estimation. Methods based on amplitude method, short-time Fourier transform method, and power spectrum entropy method are used to binarize the output of the Duffing oscillator array, and their performance is compared. The pseudo-random sequence is estimated using Duffing oscillator array synchronization, and the carrier frequency parameters are obtained by the relational expressions and characteristics of the difference frequency. Simulation results show that this blind estimation method overcomes limitations in prior knowledge and maintains good parameter estimation performance up to an SNR of-35 d B.     

基于神经网络的航天光学遥感器在轨信噪比测试方法

提出一种基于神经网络的航天光学遥感器在轨信噪比的的测试方法。通过模拟得到了大量的包含有不同信噪比等级的遥感图像,并将其作为网络训练和测试的样本。通过对遥感图像进行分析,找到了分别与景物结构和噪声有关的特征向量,并将其作为神经网络的输入。在对大量样本图片进行训练后,可完成对由遥感器传输下来的任意一幅地面景物图像进行信噪比的测试,从而避免了传统方法对特定地面景物目标在成像测量中的诸多弊端,平均测量误差约为10%。     

基于四波段图像融合的彩色夜视方法研究

为了在低照度条件下获取目标的彩色图像,提高夜间对目标的识别能力,提出了基于四波段图像融合的彩色夜视方法.采用F-P滤光片设计出了透射中心在三基色光中心波长及近红外波段的四波段滤光片,在各个透射区域的平均透射率均达90%以上;将四波段滤光片设计成圆形滤色轮结构,用分光计测得了各波段滤光片的光谱透射性;对加入滤光片后的系统信噪比进行了分析和计算,分析结果表明加入滤光片不会引入噪声,计算结果表明加入蓝色、绿色、红色和近红外波段滤光片的成像系统的信噪比分别是原来单色微光夜视系统信噪比的19.59%、38.45%、47.28%和46.70%.借助国产超二代像增强器在微光实验室进行了四波段图像采集及彩色图像融合实验,实验时光照度分别为1×10~(-3)lx和1×10~(-1)lx,对获取的图像质量进行了评价.结果表明:在照度为1×10~(-3)lx时,融合的彩色图像在均值、方差和熵这三项指标上均优于过滤后的蓝色和绿色的单色图像,且由于彩色图像中利用了近红外图像进行增强,使得彩色图像亮度更高,颜色分辨性更好;在照度为1×10~(-1)lx时,融合后的彩色图像的信息熵比红、绿、蓝三种基色图像的大,彩色图像携带的信息量更大.本文的研究对彩色夜视成像系统的设计和研发具有借鉴和指导意义.     

Deep learning for image reconstruction in thermoacoustic tomography

Microwave-induced thermoacoustic tomography(TAT)is a rapidly-developing noninvasive imaging technique that integrates the advantages of microwave imaging and ultrasound imaging.While an image reconstruction algorithm is critical for the TAT,current reconstruction methods often creates significant artifacts and are computationally costly.In this work,we propose a deep learning-based end-to-end image reconstruction method to achieve the direct reconstruction from the sinogram data to the initial pressure density image.We design a new network architecture TAT-Net to transfer the sinogram domain to the image domain with high accuracy.For the scenarios where realistic training data are scarce or unavailable,we use the finite element method(FEM)to generate synthetic data where the domain gap between the synthetic and realistic data is resolved through the signal processing method.The TAT-Net trained with synthetic data is evaluated through both simulations and phantom experiments and achieves competitive performance in artifact removal and robustness.Compared with other state-of-the-art reconstruction methods,the TAT-Net method can reduce the root mean square error to 0.0143,and increase the structure similarity and peak signal-to-noise ratio to 0.988 and 38.64,respectively.The results obtained indicate that the TAT-Net has great potential applications in improving image reconstruction quality and fast quantitative reconstruction.     

Robust image encryption by combined use of integral imaging and pixel scrambling techniques

A robust image encryption method by using the integral imaging and pixel scrambling (PS) techniques is proposed. In this method, pixels of the cover image are scrambled with the PS technique and elemental images for this scrambled image are picked up through a lenslet array. Subsequently, an encrypted image is obtained by scrambling these picked-up elemental images. Since this encrypted image has the hologram-like property of data redundancy resulted from the integral imaging scheme, while it can as well be decoded by multiple keys such as the orders of pixel scrambling and the pickup conditions of the elemental images, its security against the various attacks could be dramatically improved. Good experimental results also confirm that the proposed method could provide more enhanced robustness against data loss and Gaussian noises compared to the conventional methods.     

An improved automatic time-of-flight picker for medical ultrasound tomography

Objective and motivationTime-of-flight (TOF) tomography used by a clinical ultrasound tomography device can efficiently and reliably produce sound-speed images of the breast for cancer diagnosis. Accurate picking of TOFs of transmitted ultrasound signals is extremely important to ensure high-resolution and high-quality ultrasound sound-speed tomograms. Since manually picking is time-consuming for large datasets, we developed an improved automatic TOF picker based on the Akaike information criterion (AIC), as described in this paper.MethodsWe make use of an approach termed multi-model inference (model averaging), based on the calculated AIC values, to improve the accuracy of TOF picks. By using multi-model inference, our picking method incorporates all the information near the TOF of ultrasound signals. Median filtering and reciprocal pair comparison are also incorporated in our AIC picker to effectively remove outliers.ResultsWe validate our AIC picker using synthetic ultrasound waveforms, and demonstrate that our automatic TOF picker can accurately pick TOFs in the presence of random noise with absolute amplitudes up to 80% of the maximum absolute signal amplitude. We apply the new method to 1160 in vivo breast ultrasound waveforms, and compare the picked TOFs with manual picks and amplitude threshold picks. The mean value and standard deviation between our TOF picker and manual picking are 0.4 μs and 0.29 μs, while for amplitude threshold picker the values are 1.02 μs and 0.9 μs, respectively. Tomograms for in vivo breast data with high signal-to-noise ratio (SNR) (∼25 dB) and low SNR (∼18 dB) clearly demonstrate that our AIC picker is much less sensitive to the SNRs of the data, compared to the amplitude threshold picker.Discussion and conclusionsThe picking routine developed here is aimed at determining reliable quantitative values, necessary for adding diagnostic information to our clinical ultrasound tomography device - CURE. It has been successfully adopted into CURE, and allows us to generate such values reliably. We demonstrate that in vivo sound-speed tomograms with our TOF picks significantly improve the reconstruction accuracy and reduce image artifacts.