Mobile QR Code QR CODE

Object-wise Secure Image Display Method for Screen Capture Protection

(Jinwoo Kang) ; (Sang-ug Kang)

In this study, we propose an image copyright protection technology to prevent analog hole attacks using a deep object detection network. Instead of applying a virtual fence to the whole image, only a region of copyright is protected with fences. The region of copyright interest is called the protection zone and is generated by merging bounding boxes found with a YOLO network. The protection zone is adjusted to satisfy the fence generation rules, and then the virtual fence is finally applied. The main goal of this study is to attenuate the image degradation caused by virtual fences while preserving copyright protection capability. Since the virtual fence method uses the afterimage effect of the human visual system, it is necessary to generate images perceived by a human eye at every moment in order to calculate the objective image quality compared to the original unprotected image. Perceived images are generated at every 0.001 seconds using the mathematical model of the afterimage effect. The protection zone is found using object bounding boxes detected using a YOLO deep neural network. We applied the proposed method to four animation images, and the simulation results show that it achieves image quality that is about 3dB better than that of other methods.

Face Anti-spoofing: A Comparative Review and Prospects

(Wonjun Kim)

Face recognition and verification have been widely used for user authentication on various devices. To protect personal biometric information, face anti-spoofing techniques have been explored based on learning of the properties of various spoofing attacks against real faces. However, the direction of these studies has now encountered an important issue of validation for unseen types of spoofing attacks. Most methods have focused on revealing the differences between fake and real faces in a given dataset, but the performance significantly drops when such methods are tested on unseen samples that are different from those of the dataset used for training models. To cope with this limitation, several researchers have started to generalize the feature space across different datasets (i.e., free to the domain property). The goal of this paper is to provide a comprehensive review of face anti-spoofing methods with a systematic taxonomy, methodologies, and constructive prospects.

End-to-end Automatic Sleep Staging Algorithm using Convolution Neural Network and Bidirectional LSTM

(Jaewoo Baek) ; (Suwan Baek) ; (HyunSu Yu) ; (JungHwan Lee) ; (Cheolsoo Park)

In order to measure sleep quality, sleep experts manually classify sleep stages through polysomnography (PSG) signals. However, it is time-consuming and labor-intensive work. Thus, automatic sleep stage classification methods are needed. In this study, we propose an end-to-end automatic sleep staging algorithm using a one-dimensional convolutional neural network (1D-CNN) based on an inception network and bidirectional long short-term memory (bi-LSTM). First, a feature map was extracted from input data using the 1D-CNN architecture without preprocessing. Secondly, bi-LSTM learned a stage transition rule using the feature maps. In addition, we used the sleep-EDF public dataset to evaluate our model, and only one channel of EEG signal was used to save computational cost. The accuracy and macro-averaged F1 score of the classification performance were 85.05% and 79.05%, respectively. These results demonstrate state-of-the-art performance compared to previous studies using the same dataset, yielding an effective method for an automatic sleep staging algorithm.

Mining for Building Energy-consumption Patterns by using Intelligent Clustering

(Nam Anh Dao) ; (Hai Minh Nguyen) ; (Khanh Tung Nguyen)

We present a method for the computational problem of mining for the energy-consumption patterns of apartments in residential buildings. We show a consistent scheme for how to apply data mining in order to discover partitions that constitute electrical consumption. The method is geared to stabilize robust learning and prediction by combining cluster analysis of time-series data and iterative gradient boosting from auto-regression in learning. Together with data preparation, such as the analysis of time-series patterns and well-formulated features, clustering methods can be used to specify group-based energy consumption data. Hence, we propose to use k-Means and agglomerative clustering, which adapt to the time-series data for grouped apartments. Then, robust gradient boosting is implemented to predict the levels of energy consumption for each group. Finally, prediction of energy consumption for the whole building is estimated. Our experimental evaluation demonstrates that the method allows significantly fewer errors than previous techniques.

Comparison of Program Representations on Vulnerability Detection with Graph Neural Networks

(Yoola Choi) ; (Young-Woo Kwon)

As software vulnerabilities have surged, efforts to discover them have increased. The syntactic and semantic information of a program is required to detect vulnerabilities. Each information can be represented as a graph, such as Abstract Syntax Tree and Program Dependency Graph. In this paper, the program representations were extracted using various static analysis tools, including Clang Static Analyzer, Joern, and SVF, and compared using Graph Neural Networks to select the appropriate representations for vulnerability detection in C/C++. From the comparison, PDG shows the best performance among the multiple representations. This result indicates a suitable program representation and a tool for vulnerability detection that can be utilized in research utilizing graph neural networks.

Effective Electricity Demand Prediction via Deep Learning

(Daegun Ko) ; (Youngmin Yoon) ; (Jinoh Kim) ; (Haelyong Choi)

Prediction of electricity demand in homes and buildings can be used to optimize an energy management system by decreasing energy wastage. A time-series prediction system is still a challenging problem in machine learning and deep learning. Our main idea is to compare three methods. For this work, we analyzed an electricity demand prediction system using the current state-of-the-art deep-learning methods with a machine-learning method: error correction with multi-layer perceptron (eMLP) structure, autoregressive integrated moving average (ARIMA) structure, and a proposed structure named CNN-LSTM. For this, we measured and collected electricity demand data in Germany for home appliances. We report the prediction accuracy in terms of the mean square error (MSE) and mean absolute percentage error (MAPE). The experimental result indicates that CNN-LSTM outperforms eMLP and ARIMA in accuracy.

Wide Tuning-range LO Chain with a Hybrid Inductor for LTE and SAW-less GSM Transceiver

(Seonghan Ryu)

A low phase noise, multi-standard local oscillator (LO) chain for LTE and GSM is presented. The LO chain is composed of a dynamic divider and a CMOS VCO with a hybrid inductor using both bondwire and a planar spiral inductor in the same area. This approach improves both tuning range and phase noise characteristics without additional area loss. Based on this concept, the LO chain with a hybrid inductor VCO is implemented in a 65nm CMOS process. The measured performance satisfies wide tunability and most stringent GSM transmitter phase noise specification, which proves that the LO chain can be used for LTE and SAW-less GSM transceiver.

Study of Battery State-of-charge Estimation with kNN Machine Learning Method

(Teressa Talluri) ; (Hee Tae Chung) ; (Kyoojae Shin)

Electric vehicles have high demand due to their ecofriendly nature. From this point of view, lithium batteries have gained high attention in recent days due to their high efficiency and long life time. Hence, it is of the utmost importance to evaluate the battery characteristics, such as the state of charge (SOC), depth of discharge (DOD), and remaining life of a battery to ensure battery safety. These parameters were derived in order to estimate the battery life time before degradation. This estimation is very much required in making a decision about battery usage in the future. In this study, the SOC of a lithium polymer battery was evaluated in a real-time experiment. Charging and discharging cycles were done, and we obtained the voltage, current, and time data from the experimental result. This experimental data trained machine learning methods such as the kNN (k Nearest Neighbor) method to estimate the SOC more precisely. After training the model, a test was done. The proposed estimator was calibrated by experimental data. The results are satisfactory with accuracy of 98% and mean absolute error (MAE) as low as 0.74[%].

Design of Temperature Control Battery Pack of Electric Vehicle using PCM Thermal Characteristics

(Kyoojae Shin)

This paper is about the thermal management of an electric vehicle battery pack with phase change materials. The battery pack safe operating temperature is 25 to 40°C. Under extreme conditions, the battery pack of electric vehicles will generates high heat during operation, so for this we proposed a thermal management of with phase change materials(pcms). At high heat genartion conditions, pcms will observe heat from battery pack. In addition, in extreme cold temperatures of zero degrees pcm will give warmth to battery pack. In this study, we analyzed the PCM properties and evaluated the hysteresis properties for effective heating and cooling of the battery pack. In order to perform optimal temperature control of the battery pack, a BTMS was designed and fabricated using the thermal characteristics of EG28 and EG26 PCMs and an optimized hysteresis PID control method. The proposed temperature control system was confirmed to be excellent through experiments.

A Novel Computation Method for Tag Estimates in DFSA-based RFID Systems

(Young-Beom Kim)

In this paper, we propose a novel computation method for tag estimation in the dynamic framed slotted ALOHA (DFSA) protocol, based on Vogt’s estimate (VE). The estimation target is system load λ and tag estimate N? is computed from the estimated values of λ, instead of the direct approach taken in the existing method. Starting from the definition of VE, we derive a single concise equation for the estimate. For the numerical solution, we present a load estimation algorithm, which turns out to be computationally very efficient. The computational complexity is roughly O(10 log10 Nmax), in contrast to O(Nmax) in the existing method for maximum possible number of tags, Nmax. Through computer simulations, we show that the proposed method indeed yields estimation results almost identical to those by the existing method in most cases.