Search

The microwave scatterometer onboard the China-France Oceanography Satellite (CFOSAT Scatterometer, CSCAT) is the first rotating fan-beam scatterometer (RFSCAT) in operation. It is used to obtain the ocean surface wind field by measuring the backscattering coefficient over a large observation swath. To estimate the backscattering coefficient, the scatterometer needs to measure the scattered signal accurately. The process of removing noise from the echo signal (scattered signal plus noise) is called noise subtraction. Noise subtraction has a great influence on the measurement of the backscattering coefficient and subsequent wind field retrieval. Therefore, accurate noise subtraction is necessary and important. In this article, an improved noise subtraction method for CSCAT is proposed. According to the distribution characteristics of backscattering coefficients, noise correction factors are introduced to improve the noise subtraction of CSCAT. The ocean surface wind fields are obtained using the improved backscattering coefficient data, and the deviation of wind speed and direction are calculated by comparison with the European Centre for Medium Range Weather Forecasts wind fields. The results show that the improved noise subtraction method can lead to improvement of the quality of wind field effectively. In addition, based on the improved method of noise subtraction, a method of estimating the noise correction factor is proposed at a large incidence angle, furtherly improving the noise subtraction method. The results show that the proposed method is effective.

With the augmentation of the non-zero conjugate correlation statistics of some modulating signals, enlarged degrees of freedom (DOFs) can be achieved with conventional sparse arrays as proposed by many literatures. But little research has been carried out on sparse array design with the augmented statistics. In this paper, a novel sparse array geometry which exploits both the correlation and conjugate correlation statistics is proposed. It stems from the prototype nested array with transposed subarrays. The sensor locations are determined systematically, and the closed-form expression for the achieved aperture under given number of physical sensors is derived. In our scheme, a new hole-free virtual array called the sum difference coarray is constituted with the proposed sparse array, in which all the lags are consecutive. Due to the output of the coarray is obtained by averaging the repeated lags generated by the addition and subtraction of the physical sensor location, the hole-free structure of the coarray enables all the lags engaged in averaging and makes the output of the coarray more stable. Moreover, the coarray has larger aperture than existing sparse arrays, so that bringing more degrees of freedom. In the end, simulations are conducted to demonstrate the superior performance of the proposed scheme over other existing structures.

Electromyographic (EMG) measurements of the respiratory muscles provide a convenient and noninvasive way to assess respiratory muscle function and detect patient activity during assisted mechanical ventilation. However, surface EMG measurements of the diaphragm and intercostal muscles are substantially contaminated by cardiac activity due to the vicinity of the cardiac muscles. Many algorithmic solutions to this problem have been proposed, yet a conclusive performance comparison of the most promising candidates currently is missing. The objective of this work is to provide a quantitative performance comparison of six previously proposed cardiac artifact removal algorithms operating on single-channel EMG measurements, and two newly proposed, improved versions of these algorithms. Algorithmic performance is evaluated quantitatively based on four different measures of separation success, using both synthetic validation signals and electromyographic measurements of the respiratory muscles in eight subjects. The compared algorithms are two versions of the empirical template subtraction algorithm, two model-based Bayesian filtering procedures, a wavelet denoising approach, an empirical mode decomposition (EMD) based approach, and classical high-pass filtering. Different algorithms perform well with respect to different performance measures. Template subtraction algorithms yield the best results if the characteristics of the raw signal are of interest, while filtering algorithms like simple high-pass filtering, wavelet denoising, and EMD-based denoising show superior performance for calculating a cleaned envelope signal. No algorithm completely removes the cardiac interference, but the characteristic errors introduced by the considered algorithms differ. Hence, the choice of the algorithm to use should be made depending on the target application. Finally, we also demonstrate that our empirical SNR measure, which can be calculated without knowledge of the true, undisturbed signals, correlates strongly with the exact reconstruction error. Thus, it represents a reliable indicator for algorithm performance on real measurement data.

Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) has been increasingly used for lesion detection in breast cancer diagnosis for its capability to provide spatial-temporal information. However, the massive and complex 4D spatial-temporal DCE-MRI data make the diagnosis process lengthy and error-prone. Moreover, normal fibroglandular tissue is occasionally enhanced through background parenchymal enhancement (BPE), which can degrade the performance of current algorithms. We propose a new method using a 3D Clifford analytic signal (CAS) approach for breast lesion segmentation of DCE-MRI data. A 2D temporal image is constructed from all the 2D DCE-MRI slices at different scanning time points on a given transverse plane, according to the CAS approach. Then, a 3D Clifford temporal image (CTI) is constructed by successively stacking temporal images. The proposed CTI can distinguish lesion regions both visually and quantitatively compared to the traditional DCE-MRI subtraction image. Finally, we employ a fully convolutional network (FCN) model for breast lesion segmentation using the CTI as one of the inputs. Experimental results on an independent public dataset (TCIA QIN breast DCE-MRI) and a private household breast DCE-MRI dataset (TBD) show that the proposed method can achieve superior performance over current methods, both qualitatively and quantitatively.

The fundamental goal of augmented reality (AR) is to integrate virtual objects into the user’s perceived reality seamlessly. However, various issues hinder this integration. In particular, Optical See Through (OST) AR is hampered by the need for light subtraction due to its see-through nature, making some basic rendering harder to realize. In this paper, we realize mutual shadows between real and virtual objects in OST AR to improve this virtual-real integration. Shadows are a classic problem in computer graphics, virtual reality, and video see-through AR, yet they have not been fully explored in OST AR due to the light subtraction requirement. We build a proof-of-concept system that combines a custom occlusion-capable OST display, global light source estimation, 3D registration, and ray-tracing-based rendering. We will demonstrate mutual shadows using a prototype and demonstrate its effectiveness by quantitatively evaluating shadows with the real environment using a perceptual visual metric.

Contrast-enhanced ultrasound (CEUS) is a vascular imaging technique that has several clinical and preclinical applications. It is generally performed in 2-dimensional (2-D) space, which is not ideal for imaging complicated blood vessel networks. Alternatively, volumetric data can be acquired at a high frame rate with 2-D matrix arrays, although the image quality is reduced. Alternatives to conventional delay-and-sum (DAS) beamforming can be used to compensate for the poor resolution and high side lobe levels. The purpose of this research was to determine if the recently developed null subtraction beamforming technique can improve vessel resolution during volumetric CEUS imaging. A tissue-mimicking phantom with a 2.5 mm vessel lumen was perfused with a microbubble contrast agent and imaged with a programmable US scanner (Vantage 256, Verasonics Inc) equipped with a 1024-element matrix array transducer (7.5 MHz, Vermon). Volumetric data was acquired and beamformed with both null subtraction and conventional DAS. The resulting temporal maximum intensity projections were compared. Qualitatively, the vessel lumen was clearer and side lobe noise was greatly reduced by null subtraction beamforming. The signal-to-noise ratio (SNR) of the vessel was also improved from 9.8 dB for DAS to 13.3 dB for null subtraction. Null subtraction beamforming is a simple technique that improves the quality of volumetric CEUS images acquired using a 2-D matrix array transducer.

There are two problems in SAR image change detection when using difference images (DIs), i.e., 1) the subtraction operation results in serious loss of semantic information in DIs; and 2) the boundary of DI is uncertain. We propose a change detection method based on Neural-CRF structure. Firstly, Transformer-UNet (TR-UNet) is designed to provide the unary potential for CRF. The TR-Attention module improves the semantic expression ability of UNet by introducing the multi-head attention mechanism of TR. Secondly, a cascade CRF as Recurrent Neural Network, named as C-CRF-RNN, is proposed to update the unary potential and pairwise potential simultaneously. This improves the ability of CRF-RNN to refine pixel-level label prediction. Experiments show that the proposed method consistently outperforms the state-of-the-art methods on two benchmarks including berne data and ottawa data.

In this paper, we present spiking neural network (SNN) conversion technique optimized for converting low bit-width artificial neural networks (ANN) trained with quantization aware training (QAT). Conventional conversion technique suffers significant accuracy drop on QAT ANNs due to different activation function used for QAT ANNs. To minimize such accuracy drop of the conventional conversion, the proposed technique uses Spike-Norm Skip, which selectively applies threshold balancing. In addition, subtraction based reset is used to further reduce accuracy degradation. The proposed conversion technique achieves an accuracy of 89.92% (0.68% drop) with a 5-bit weight on CIFAR-10 using VGG-16.

We propose a high-rate generation method of optical Schrodinger cat states [1] . More than a decade, the standard method of cat-state generation has been photon subtraction from squeezed vacuum states [2] . Its constraints on experimental parameters, however, limit the generation rate. We overcome the problem by considering photon-number measurement in one mode of arbitrary two-mode Gaussian states, which we call "generalized photon subtraction". Analysis of such a general situation is challenging, but we analytically derive an efficient condition by describing the state-generation process using wave functions. Supposing realistic experimental conditions, the generation rate of cat states with large-amplitude (| α| ≥ 2) is improved by multiple orders from the conventional photon subtraction. Our proposal would open a way for quantum error correction and coherent-state-based quantum computing.

In indoor environment that elderlies live, ultrasonic array sensors which are believed to protect their privacy, are hidden in the walls for collecting distance signals of all things in the place. The distance signals in term of array which are assumed as distance images are then needed to process for detecting elderly fall. This paper discusses how to process these distance images for fall detection by using machine learning. First of all, moving objects in the indoor environment are detected by subtraction between neighboring frames, all continuing frames of moving objects are classified by machine learning, and classification results of all frames are finalized to recognize elderly fall. The experiments with some samples and constructed prototype based on the fall standard are performed, and the results reveal good performance in accuracy.

Currently, hard disk drive has been struggled to overcome the super-paramagnetic limit that restricts the density increment. Bit-patterned magnetic recording (BPMR) is a candidate that can increase an areal density (AD) up to 4 Tb/in2. However, inter-track interference (ITI) arising from a narrow track width at high AD severely degrades the system performance. This study proposes the ITI subtraction technique with turbo iteration in a coded BPMR system. This method refines the equalized sequence by subtracting the ITI using an imitated ITI sequence that is generated by the soft information obtained from turbo iteration and ITI coefficients. The refined sequence that contains the partial ITI is sent to the turbo iteration as many rounds as needed. Simulation results indicate that the proposed system outperforms the conventional system whether or not media noise is considered.

In this paper, the classic template matching sequential similarity detection algorithm(SSDA) has a large amount of calculation and slow operation time, and an improved SSDA is proposed. By introducing the integral image, the point-by-point summation in the matching template window is transformed into the addition and subtraction of four positions in the integral image, which reduces the amount of calculation for point-by-point accumulation and summation. The laser image and the classic lena image are used as the experimental images to be matched, and the image with gray scale changes and noise added as the template image is used for image matching. Compared with the traditional normalized gray cross-correlation(NCC) algorithm and the improved NCC algorithm for integrating images, the experimental results show that the improved algorithm can shorten the running time, have a certain anti-interference ability against noise, and achieve image matching.

Local fields potentials (LFP) can be contaminated by different internal and external sources of noise during their recordings. In cases where artefacts are present in them, automatic detection tools are needed to speed up the high accuracy detection process, followed by their removal to successfully use these recordings. This process is facilitated by a pool of supervised machine learning based tools which require labelled data for training. These algorithms have the capacity to distinguish between normal brain patterns and artefacts from an individual or a group, which is more flexible than template matching and subtraction. In addition, their portability has seen developments from both the software and hardware perspective. For many tools, LFP signal power is used as a gauge to measure and eventually label the artefacts portion of the LFP. This work explores how signal power affects the detection and classification accuracy of artefacts. Results show that a higher threshold value impacts positively the accuracy, due to less false positives in the data, without compromising the specificity in balanced datasets.

An automatic video matting method based on unsupervised matting using edge feature is proposed in this paper. Extract foreground and background objects from video sequences is an essential task as video matting. But it needs an automatic process to determine foreground and background area so the human effort can be reduced. We proposed a method to decide mechanical constraints using edge features on video matting. Edge features are used to find a gradient vector from each RGB color channel to obtain automatic scribbles. Comparing the results, alpha matte from our proposed method with ground truth has less than 0,01 in the MAE score.

A system using computer vision techniques for tracking and providing early information of hazardous locations in huge gatherings is the need of the hour. Also, the number of video streams generated are huge and are challenging to watch them personally. The proposed system is based on the Optical Flow based estimations and detects sequences of crowd motions that are characteristic for devastating congestions. Initially, the temporal features of the scenes are extracted using Motion History Image (MHI) technique. MHI technique involves the weighted subtraction of consecutive frames of the video stream. Then the Optical Flow vectors are calculated using the Lucas-Kanade method. Segmentation of Optical Flow fields is done, and hence directional histograms of motion magnitude against motion direction are determined for respective segments. Their entropy and absolute bin magnitudes characterize the graphs. Thresholds are chosen such that there is a demarcation between sparsely crowded and densely crowded segments in the frames. Localization of crowd density levels as unpopulated, sparse and dense helps in providing immediate attention to critical areas of congestion. The directional analysis helps in having knowledge of the significant direction of motion in sparsely crowded regions.

Contrast-enhanced Computed Tomography (CT) imaging is very helpful for the detection of tumor metastasis and cancer cells. It is one of the most effective means of clinical imaging examination. The objective of this paper is to realize the quantitative reconstruction of two contrast agents at the same time and obtain the non-contrast-enhanced images in four energy bins as an additional product. This paper presents an iterative material decomposition method based on volume conservation and K-edge characteristics. Compared with other K-edge based material decomposition algorithms, the proposed method can simultaneously quantify the two contrast agents and make better use of the energy information provided by the photon-counting detector. The proposed algorithm was compared with the image-domain K-edge subtraction, Angular Rejection and filtered back projection (FBP) reconstruction. Numerical simulation and Monte Carlo simulation were used to verify the effectiveness of the proposed algorithm. We found that the proposed algorithm has higher efficiency and is meaningful for the quantification of contrast agent concentration. The proposed algorithm can achieve the quantitative separation of dual K-edge contrast agent by a single scan, which is significant for reducing the radiation dose to patients and improving the efficiency of material decomposition.

Facial attribute editing aims to modify face images in the desired manner, such as changing hair color, gender, and age, adding or removing eyeglasses, and so on. Recent researches on this topic largely leverage the adversarial loss so that the generated faces are not only realistic but also well correspond to the target attributes. In this paper, we propose Residual Attribute Generative Adversarial Network (RAG), a novel model to achieve unpaired editing for multiple facial attributes. Instead of directly learning the target attributes, we propose to learn the residual attributes, a more intuitive and understandable representation to convert the original task as a problem of arithmetic addition or subtraction for different attributes. Furthermore, we propose the identity preservation loss, which proves to facilitate convergence and provide better results. At last, we leverage effective visual attention to localize the related regions and preserve the unrelated content during transformation. The extensive experiments on two facial attribute datasets demonstrate the superiority of our approach to generate realistic and high-quality faces for multiple attributes. Visualization of the residual image, which is defined as the difference between the original image and the generated result, better explains which regions RAG focuses on when editing different attributes.

To potentially overcome the practical security loopholes of CV-QKD protocols, in this paper, we propose to use the optimized eight-state measurement-device-independent (MDI) protocol and demonstrate that it can significantly outperform corresponding Gaussian modulation-based MDI and virtual photon subtraction-based MDI CV-QKD protocols in terms of both secret-key rate and achievable transmission distance. Contrary to the common belief that virtual photon subtraction method can extend the distance of MDI CV-QKD protocols, we show that this is not true for fully optimized MDI CV-QKD protocols and realistic system parameters.

We present a beat-to-beat heart rate tracking algorithm that is designed especially to handle the nonstationary motion artifacts often encountered using photoplethysmographic (PPG) signals acquired from smartwatches or a forehead-worn device, during intense exercise. To date, many algorithms have been based on tracking heart rates during intense exercise using an 8-second average of heart rates, which does not accurately capture the large variation in instantaneous heart rates during exercise. In this paper, we propose a novel technique that can accurately estimate heart rates from wearable PPG signals with subjects running on a treadmill and making other sudden movements. The proposed algorithm includes three parts: 1) time-frequency spectrum estimation of PPG and accelerometer signals, 2) motion artifact removal by subtraction of the time-frequency spectra of the accelerometer signals from the PPG signals, and 3) postprocessing to further reject motion artifact-affected heart rates followed by interpolation of removed heart beats using a cubic spline approach. The proposed approach was compared to one of the recent and most accurate algorithms. The results of the proposed and compared algorithms were evaluated with two datasets (IEEE Signal Processing Cup (N=12) and our own dataset (N=10)) obtained from a smartwatch and a forehead PPG sensor with subjects running on a treadmill. The reference heart rates were obtained from a chest-worn ECG. Our method, using a 12 second windowed segment, resulted in an average absolute error of only 2.94 beats per minute and an average relative error of 2.42 beats per minute, which are a 71% and 94% improvement, respectively, over the compared algorithm.

ICT as educational tools are very beneficial for the teacher and for students in the 2nd grade of the primary level since through the GCompris software. The purpose of this research is to analyze how students in the second grade of primary school implement the GCompris program as a teaching resource in the area of mathematics to solve problems of addition, subtraction and mathematical language. The sample was of 36 students were used as an experimental group (pre-test - post-test), the participants were between 8 and 9 years old. Finally, an exit evaluation was applied to compare the result and see if the software was relevant to improve its mathematical performance. Finally, the results showed that through the comparative chart students managed to improve their mathematical performance.

Addition calculation and subtraction calculation are the most basic and common arithmetic operations. Different computational equations use different strategies, and the range and intensity of brain regions are also different. The research on cognitive processing strategies of addition calculation and subtraction calculation is helpful to explore the brain activity regularity of different cognitive processing strategies of human brain, provide reference for the further development of artificial intelligence and computer intelligence, and put forward cognitive model to realize the research on it when it is difficult to obtain a certain cognitive task data.

In this research, we analyze a very common issue called wandering baseline artifact (WBA) that has with electrocardiogram (ECG) signals. This artifact is due to normal patient's movements and is present on each derivation. This wandering baseline has a very low frequency, normally under 0.2 Hz. The wandering baseline on the ECG signals produce an alteration on the ‘P’ segment (the information of the atrial depolarization) and the ‘ST’ segment (the information of a ventricular contraction). We have developed the linear filter that has used to remove this WBA by keeping the nature of the original ECG signals and making sure not to vanish clinical important information from the original signals. In this research, we used ECG signals provided by the Physionet, which is a database website that provide physiological signals for research purposes. The filters (median and moving average) that used in the proposed algorithm is to obtain the signal that we do not want from our original signal. Finally, we do a subtraction between the original signal and the filtered signal and analyze both signals (input and output) in the frequency domain using the Welch Periodogram that indicate the reduction of wandering baseline since the power density spectrum decrease among low frequencies.

The single-image super-resolution (SISR) aims to recover a potential high-resolution image from its low-resolution version. Recently, deep learning-based methods have played a significant role in super-resolution field due to its effectiveness and efficiency. However, most of the SISR methods neglect the importance among the feature map channels. Moreover, they can not eliminate the redundant noises, making the output image be blurred. In this paper, we propose the residual subtraction and attention network (RSAN) for powerful feature expression and channels importance learning. More specifically, RSAN firstly implements one redundance removal module to learn noise information in the feature map and subtract noise through residual learning. Then it introduces the channel attention module to amplify high-frequency information and suppress the weight of effectless channels. Experimental results on extensive public benchmarks demonstrate our RSAN achieves significant improvement over the previous SISR methods in terms of both quantitative metrics and visual quality.

In this paper, a neural approach based on using Long-Short Term Memory (LSTM) neural networks is proposed to diagnose patients suffering from PD. In this study, it is shown that the temporal patterns of the gait cycle are different for healthy persons and patients. Therefore, by using a recurrent structure like LSTM, able to analyze the dynamic nature of the gait cycle, the proposed method extracts the temporal patterns to diagnose patients from healthy persons. Utilized data to extract the temporal shapes of the gait cycle are based on changing vertical Ground Reaction Force (vGRF), measured by 16 sensors placed in the soles of shoes worn by each subject. To reduce the number of data dimensions, the sequences of corresponding sensors placed in different feet are combined by subtraction. This method analyzes the temporal pattern of time-series collected from different sensors, without extracting special features representing statistics of different parts of the gait cycle. Finally, by recording and presenting data from 10 seconds of subject walking, the proposed approach can diagnose the patient from healthy persons with an average accuracy of 97.66%, and the total F1 score equal to 97.78%.

Chest radiographs are widely used in clinic, typically in diagnosis and treatment of lung tumors. Accurate registration of chest radiographs is surprisingly important by providing correspondence between different images, which is however still a challenge at present. To solve this problem, we propose a diffeomorphic demons based algorithm for registration of chest radiographs in this paper. Specifically, an improved similarity metric is designed in terms of a log Euclidean formulation and solved by a diffeomorphic framework with composition of two stationary velocity fields, which averts heavy computation that often occurs in spatial transformation of the traditional log demons method. To evaluate the performance of the proposed algorithm, qualitative and quantitative validation methods are carried out and subtraction images are generated. Results show that the proposed diffeomorphic demons based algorithm obtain more accurate registration of chest radiographs compared with active demons, additive demons, and traditional log demons methods.

This paper proposes a method for automatic generation of the power network diagram of plants and substations for grid dispatching system. Establish a local graphic feature library in layers, extract the feature quantity of the existing plant diagrams based on graph-model-data-database integration. Match the features of new diagram and the existing diagrams layer by layer. Then Generate the new diagram automatically with combination, replacement, addition and subtraction of the existing diagrams containing the characteristics of the target diagram.

A multiplicative inverse (MI) algorithm is used in several fields, like cryptography algorithms. There are many MIs, however, these algorithms suffer from using several multiplication and division operations, which take more execution time than additions and subtractions. We created a new algorithm called Clockwise-Displacement (CDMI) by using addition and subtraction operations in the iterative steps instead of multiplications and divisions. Additionally, numerous MIs face the undecidable problem because of the floating-point issue. Whereas, CDMI tackles this issue by converting the domain from the floatingpoint space to integer space. Therefore, CDMI declines the time consuming to calculate the multiplicative inverse by applying fewer divisions and multiplications (expensive operations) and addresses the rounding error issue in some MI Algorithms.

The use of cloud technologies for processing confidential data requires a solution to the data security problem. One of the mechanisms to solve it is homomorphic encryption. However, homomorphic encryption only allows the arithmetic operations of addition and multiplication to be performed over encrypted numbers. Consequently, when implementing algorithms for matrix algebra, artificial neural networks, and deep learning, it becomes necessary to implement the comparison operation in a homomorphic cipher. In this paper, we study methods for comparison operations in a homomorphic cipher. One of the methods is to use the subtraction of numbers and determine the sign of a number. Two families of polynomials and their composition are used to approximate the sign function. We provide estimates of the accuracy of the sign function approximation obtained as a result of modeling and show that they are 1.98 times better than the state-of-the-art theoretical estimate for polynomials $f_{n}(x)$ . Also, we show that the theoretical estimate of $g_{n}(x)$ is not applicable for $n\leq 4$ .

paper presents some simple dependencies, using which you can visually check the correctness of the solution of the calculation of multi-digit operations of addition, subtraction, multiplication, multiplication by module and exponentiation by module.

Basically, arithmetic entails numerals, addition, subtraction, multiplication and division. In this paper we emphasize the arithmetic abilities of the living cell pumps by presenting an abstract model for these basic operations. The verification that these arithmetic operations are working properly is done by providing a software implementation that allows to simulate how the cell pumps are doing arithmetic.

This paper suggests the efficient implementation of Modular Subtraction Unit for NTT-based polynomial multiplier on ASIC design. We will introduce the basic algorithms of Number Theoretic Transform (NTT) and the Modular Arithmetic briefly, compare the implementation of the modular subtraction unit with the one which suggested in [1] by intuitive way. We reduced about 30% transistors in 17-bit Modular Subtraction Unit.

This paper presents a compact architecture of CNN training accelerator targeted for mobile devices. Accuracy was verified using python in the CNN structure, and accuracy was compared by applying several data types to find optimized data types. In addition, floating-point operations are used in the computation of the CNN structure, and to implemented them, we have created and verified the addition, subtraction, and multiplication circuits of floating-point. The CNN architecture was verified using python, the floating point operation was verified using Vivado, and Area was verified TSMC 180nm.

In-memory-computing is an emerging approach that aims to shift computational load away from CPUs. It does so by taking over simple calculations that can be performed in memory. Apart from improving performance for these operations, this also results in a lower energy consumption, effectively rendering this technique very suitable for embedded systems.Due to being variable as well as non-volatile resistors, memristors are capable of storing analog values. This renders them particularly useful for in-memory computing.This paper presents a prototypical implementation of analog calculations (addition, subtraction and multiplication), including a way of representing the value zero. The prototype shown is based on an ESP32 microcontroller and typical calculations currently take around 1µs.

Computer vision has grown tremendously thanks to advances in convolutional neural networks (CNNs), the field of traffic sign recognition and intelligent transportation performs better because of that. However, as traffic sign recognition systems are often deployed in computing and storage constrained environments, it is critical for the network design to be accurate and real-time in these instances. Whether it is training or running an ordinary neural networks framework, it requires one or many high-power Graphic Processing Units (GPUs) and considerable memory resources. To solve this problem, this project creatively propose a 2-stage binarized multi-scale neural network framework, called B-MNN, extends Pierre Sermanet and Yann LeCun's Convolutional Networks. When training a B-MNN, we intentionally restrict the weights and activations of the model to +1 or −1. The mathematical calculation (addition, subtraction, multiplication and division) of the binarization network can be implemented using bit-wise bit operations, which is convenient for hardware to implement the algorithm, faster and more energy efficient. B-MNN approaches state-of-the-art results on the GTSRB traffic sign dataset implemented with the Keras artificial neural network library. We achieved 91.34% accuracy. Although it is slightly less accurate than models based on CNNs, the same ordinary neural network structure is an order of magnitude larger than ours. It is an important direction to realize the neural network into the realistic scene in the future.

In this paper, we introduce a method to classify animal behaviors from videos taken by a fixed-point camera. In order to classify animal behavior, it is necessary to detect and track the animals. Conventional approaches for detecting moving objects are based on background subtraction and frame subtraction. Conventional methods are not suitable for detection of animals kept indoors since they are susceptible to sunlight and shadow. We propose a method to track animals and classify their behavior using skeletal information obtained by DeepLabCut. The experimental results show that the proposed method is superior to the conventional method.

Segmentation of teeth region from the dental panoramic X-Ray images is an important task in determining various diseases. The main goal of this article is to be able to automatically segment the region of teeth in panoramic x-ray images. First, the center point of the teeth area in the images was determined automatically. Then, a feature set was developed including intensity values of pixels, x-coordinate relative to this center point, y-coordinate relative to this point, and the pixel values obtained by subtraction of maximum and minimum values in 3×3 window. CatBoost algorithm was used for machine learning. When creating the machine learning model, k-fold cross validation of training data set and grid search optimization of hyper parameters, were applied to avoid over fitting of data set. The results were analyzed using the learning curve, F1, accuracy, recall, and precision scores.

In this paper, a time-of-flight 3-D camera is presented for indoor/outdoor high-speed applications. A three-phase algorithm is proposed using one-tap/capacitor in-pixel background light subtraction. By integrating the background light only once, the standard four-phase time-of-flight operation is done in three-phases. The use of a single tap in the proposed pixel provides background light subtraction with lower mismatch and additionally, increases the camera speed. A $24 \times 24 $ pixel array image sensor, fabricated in AMS 0.35- $\mu \text{m}$ CMOS process is characterized in 2-D and 3-D mode for background light subtraction and distance estimation. The pixel pitch is $48~\mu \text{m}$ with 17.36% fill-factor. The sensor subtracts 125-klx background light with 8.2-cm distance offset at 80-cm object distance, without using an optical filter. The background-to-signal-ratio is 32.04 dB while frame-rate of the camera is 207 and 100 fps in 2-D and 3-D mode, respectively.

In this paper we present Scaled Population Subtraction to fill a void in Scaled Population arithmetic. Scaled population (SP) arithmetic is a scheme that is inspired by stochastic computing (SC), a non-conventional approximate computing method that is well known for its simplicity, area efficiency and resilience to bit errors. SP arithmetic reduces the numerical errors compared to SC and also solves the serialization limitation of SC, since it is designed to have a O(1) gate delay. Previously, SP was limited to only addition and multiplication and did not have a way to perform subtraction. This paper introduces the basic SP subtraction idea, followed by a detailed study of several ways that the basic design can be improved to reduce the computational error. Our best SP design significantly improves the error compared to our basic SP subtraction idea (reducing it by 32.3%). We also study the trade-off between design complexity of the SP subtractor against output error. Also, our implementation of the SP subtractor exhibits an improved delay, power and area compared to fixed point realizations with the same size.

This is an extension of ongoing research on Arithmetic Circuit using Fully Homomorphic Encryption using TFHE (Fast Fully Homomorphic Encryption over Torus). Just like most Homomorphic Encryption methods, the current implementation is limited to 32bit computation. Thus this was an issue that this paper attempts to resolve. This paper will focus on creating a set of arithmetic operation, namely- Addition, Subtraction and Multiplication using the logic gates provided by the library to a dynamic range to at least 128 bits. This paper will focus on creating a set of arithmetic operation, namely- Addition, Subtraction and Multiplication using the logic gates provided by the library to a dynamic range to at least 128 bits.

This is an extension of an ongoing research project on Arithmetic Circuit Homomorphic encryption (ACHE) which was implemented based on Fast Fully Homomorphic Encryption over the Torus (TFHE). This paper focuses on the integration of ASN.l Basic Encoding Rule (ITU-T x.690) into ACHE with RFC7664. Tests were performed to calculate results of 3 operand integer expressions using additional, subtraction and multiplication circuits. Next, the paper will further discuss the protocol design and the method used to achieve the calculation of expressions, and the latency of the entire process.

As a convenient and non-radiation technology, contrast-enhanced ultrasound (CEUS) has been widely used to measure blood flow and perfusion. However, when the contrast agents are flowing into the region of interest (ROI), the ultrasound can enhance all the blood vessels, so it is not nearly possible to observe a specific blood vessel and its downstream perfusion areas like digital subtraction angiography (DSA) or Arterial spin labeling (ASL). This study proposed an arterial labeling ultrasound subtraction angiography (ALUSA) based on phase-change nanodroplets (PCNDs), attempting to achieve noninvasive labeling of a specific blood vessel and ultrasound imaging of downstream blood supply and perfusion in real-time.

Single artery perfusion imaging has been an important path to study vascular diseases. Digital subtraction angiography (DSA) technology, as the gold standard for clinical vascular imaging in clinical, is invasive and radioactive. Contrast-enhanced ultrasound (CEUS) is widely used for blood pool imaging. However, the CEUS cannot estimate the structure and perfusion of a single vessel as the DSA does. This paper proposes a new method that can non-invasively label specific vessels and obtain corresponding CEUS images at the same time, which we term as the focus-flash contrast-enhanced ultrasound (FFCEUS). Through a subtraction between the normal CEUS images and the FFCEUS images, we obtain a downstream perfusion image of the labeled artery, which we termed as the ultrasound digital subtraction angiography (UDSA).

Chest radiography is one of the recommended imaging tests by the World Health Organization for childhood pneumonia diagnosis. In computer-aided diagnostic systems where radiography is the main input, its quality is crucial. The presence of foreign artifacts can, therefore, compromise the performance of these systems. In the radiography exam, foreign artifacts are very common, especially in children, due to the ingestion of objects and the need for immobilization of these patients by third parties. Identification tags, shirt buttons, catheters, tubes and in conventional scanned radiographs, fingerprints, tags, noise and inadequate brightness are some of the artifacts present. In this study, we present an efficient and very simple method for detecting and removing artifacts based on common digital image processing operations such as channel subtraction, edge detection, and morphological operations. We describe the proposed method and evaluate its performance in a database of 200 images. We show that it is robust to identify different types of artifacts regardless of their positions on the radiography. A visual inspection was used to measure the errors and the experimental results showed an accuracy of 0.98 and a processing time of about 375ms per image. As a result of this, the method demonstrates to be a very promising pre-processing tool.

In this paper, a clutter removal technique for radargrams obtained with a ground penetration radar (GPR) is presented. The technique, based in the singularity expansion method (SEM), allows to select the poles of the recorded signals that correspond to clutter. The technique was tested using a pyramidal object made of a highly reflective metallic material. An X-Y positioner that allows the GPR to perform a continuous scan was used. A direct subtraction of the previously selected clutter poles to each frame of the SAR image is performed in order to keep only the desired poles for analysis. Finally, the reconstructed radargram obtained applying the proposed technique is compared with the radargram obtained using a commonly used background-removal technique. The obtained results show that the proposed technique can be used to straightforwardly remove undesired signals measured with GPRs.

Radar-based breast imaging has shown promise as an imaging modality for early-stage cancer detection, and clinical investigations of two commercial imaging systems are on-going. Many imaging algorithms have been proposed, which seek to improve the quality of the reconstructed microwave image to enhance the potential clinical decision. However, a large artefact caused by the interface between the skin and breast interior can obscure reflections from the breast interior and, hence, tumour detection. Rotational subtraction is commonly employed to remove this artefact, however, limited work has examined the optimal parameters for rotational subtraction. In this work, the theoretical basis for rotational subtraction is examined, and the potential negative effects on image quality identified. Overall, this work suggests that it is important that the parameters of rotational artefact removal be optimised to improve radar-based breast image quality.