Mining Fine-Grain Face Forgery Cues with Fusion Modality

Face forgery detection is drawing ever-increasing attention in the academic community owing to security concerns.Despite the considerable progress in existing methods,we note that:Previous works overlooked finegrain forgery cues with high transferability.Such cues positively impact the model’s accuracy and generalizability.Moreover,single-modality often causes overfitting of the model,and Red-Green-Blue(RGB)modal-only is not conducive to extracting the more detailed forgery traces.We propose a novel framework for fine-grain forgery cues mining with fusion modality to cope with these issues.First,we propose two functional modules to reveal and locate the deeper forged features.Our method locates deeper forgery cues through a dual-modality progressive fusion module and a noise adaptive enhancement module,which can excavate the association between dualmodal space and channels and enhance the learning of subtle noise features.A sensitive patch branch is introduced on this foundation to enhance the mining of subtle forgery traces under fusion modality.The experimental results demonstrate that our proposed framework can desirably explore the differences between authentic and forged images with supervised learning.Comprehensive evaluations of several mainstream datasets show that our method outperforms the state-of-the-art detection methods with remarkable detection ability and generalizability.

Chip Layout for Adaptive Line Enhancer Design using Adaptive Filtering Algorithms and Metrics Computation for Auscultation Signal Separation

Currently,the growth of micro and nano(very large scale integration-ultra large-scale integration)electronics technology has greatly impacted biomedical signal processing devices.These high-speed micro and nano technology devices are very reliable despite their capacity to operate at tremendous speed,and can be designed to consume less power in minimum response time,which is particularly useful in biomedical products.The rapid technological scaling of the metal-oxide-semi-conductor(MOS)devices aids in mapping multiple applications for a specific purpose on a single chip which motivates us to design a sophisticated,small and reliable application specific integrated circuit(ASIC)chip for future real time medical signal separation and processing(digital stetho-scopes and digital microelectromechanical systems(MEMS)microphone).In this paper,ASIC level implementation of the adaptive line enhancer design using adaptive filtering algorithms(least mean square(LMS)and normalized least mean square(NLMS))integrated design is used to separate the real-time auscultation sound signals effectively.Adaptive line enhancer(ALE)design is imple-mented in Verilog hardware description language(HDL)language to obtain both the network and adaptive algorithm in cadence Taiwan Semiconductor Manufacturing Company(TSMC)90 nm standard cell library environment for ASIC level implementation.Native compiled simulator(NC)sim and RC lab were used for functional verification and design constraints and the physical design is implemented in Encounter to obtain the Geometric Data Stream(GDS II).In this architecture,the area occupied is 0.08 mm,the total power consumed is 5.05 mW and the computation time of the proposed system is 0.82μs for LMS design and the area occupied is 0.14 mm,the total power consumed is 4.54 mW and the computation time of the proposed system is 0.03μs for NLMS design that will pave a better way in future electronic stethoscope design.

Enhanced adaptive nonlinear extended state observer for pure feedback systems with matched and mismatched disturbances

In this paper, an enhanced adaptive nonlinear extended state observer (EANESO) for single-input single-output pure feedback systems in the presence of external time-varying disturbances is proposed. In this paper, a nonlinear system with matched and mismatched disturbances is considered. The conventional extended state observer (ESO) can only be applied to systems that are in the form of integral chains. Moreover, this method has limitations in the face of mismatched disturbances. In the presence of time-varying disturbances, the traditional ESOs cannot estimate the disturbances accurately. To overcome this limitation, an EANESO is proposed in this paper. The main idea is to design the nonlinear ESO (NESO) to estimate the states of the system and multiple disturbances simultaneously. The observer gains are considered time-varying and adjusted with adaptation laws to improve the estimation accuracy and overcome the peaking phenomenon. Next, the proposed controller is designed based on output feedback to eliminate the effects of multiple disturbances and stabilize the closed-loop system. Subsequently, the stability analysis of the closed-loop system and convergence of the observer error are discussed. Finally, the proposed method is applied to the inverted pendulum system. The simulated results show good performance of the proposed method as compared with a recently published scheme in the related literature.

Single image defogging via multi-exposure image fusion and detail enhancement

Outdoor cameras play an important role in monitoring security and social governance.As a common weather phenomenon,haze can easily affect the quality of camera shooting,resulting in loss and distortion of image details.This paper proposes an improved multi-exposure image fusion defogging technique based on the artificial multi-exposure image fusion(AMEF)algorithm.First,the foggy image is adaptively exposed,and the fused image is subsequently obtained via multiple exposures.The fusion weight is determined by the saturation,contrast,and brightness.Finally,the image fused by a multi-scale Laplacian algorithm is enhanced with simple adaptive details to obtain a clearer defogging image.It is subjectively and objectively verified that this algorithm can obtain more image details and distinct picture colors without a priori information,effectively improving the defogging ability.

Detection of transient evoked otoacoustic emissions by adaptive signal enhancement

Otoacoustic emissions (OAEs) has been considered as an excellent objective tool in clinics for diagnosing hearing loss. The signal-to-noise ratio (SNR) and correlation coefficient of OAEs are very important for the purpose of diagnosis. An adaptive signal enhancer (ASE) based on the Least Mean Square (LMS) algorithm is presented to detect transient evoked OAEs (TEOAEs). The TEOAEs detection results from 106 ears show that ASE reaches better estimation of TEOAEs than a conventional ensemble averaging (EA) technique. With the ASE, the improvement of SNR was increased faster than that with the EA and the number of sweeps required can be markedly reduced. The detection time with ASE could be shortened by about 50% in comparison with that of EA.

An improvement of adaptive line enhancer by means of coherent accumulation algorithm

It is well known that the adaptive line enhancer (ALE) is effective detector of CW signal with unknown frequency in the background of white noise. The system processing gain of ALE, when the LMS algorithm is used, however, is not satisfactory because of the presence of iterative noise and weight noise. In this paper, the coherent accumulation algorithm of ALE, called as ALECA, is suggested. It is shown that the adaptive filter employing this new algorithm possesses the ARMA structure. The experimental results also show that the processing gain of ALECA is about 14dB higher than that of conventional ALE.

Real-time adaptive noise cancelling for signal-to-noise enhancement in direct absorption spectroscopy

An adaptive filter for cancelling noise contained in the direct absorption spectra is reported. This technique takes advantage of the periodical nature of the repetitively scanned spectral signal, and requires no prior knowledge of the detailed properties of noises. An experimental system devised for measuring CH4 is used to test the performance of the filter. The measurement results show that the signal-to-noise （S/N） value is improved by a factor of 2. A higher enhancement factor of the S/N value of 5.4 is obtained through open-air measurement owing to higher distortions of the raw data. In addition, the response time of this filter, which characterizes the real-time detection ability of the system, is nine times shorter than that of a conventional signal averaging solution, under the condition that the filter order is 100.