Vol. 46, No. 12,
Dec. 2019
Digital Library
Bidirectional Warping for Single-Pass Stereo Rendering
Jaemyung Kim, Jaewon Choi, Sungkil Lee
http://doi.org/10.5626/JOK.2019.46.12.1215
Stereo images can effectively provide realism to entertainment contents such as movies and games. However, rendering for both eyes increases the computational load. Image warping can reduce number of computations required on two images to one time, but in some cases, the additional processes to fill hole that occurs during warping may require more cost than warping. In this paper, we present a bidirectional warping technique that minimizes the occurrence of hole and its implementation. The technique reduces possibility of hole creation by increasing the area of sampled pixels by changing the rendered view according to the visibility of the geometry. In a stereo image with low visual similarity, this method shows a higher quality improvement compared to the conventional image warping method. This technique is highly scalable and can be effectively used for rendering process that require complex geometric and shading calculations.
A Greedy Rule Allocation Algorithm for Efficient Distributed Complex Event Processing
http://doi.org/10.5626/JOK.2019.46.12.1222
Complex event processing (CEP) is defined as event processing for multiple stream sources to infer events that suggest complicated circumstances. As the size of stream data becomes larger, CEP engines have been parallelized to benefit from distributed computing. However, distributed CEP could duplicate redundant stream data and increase latency without consideration about the computational cost on each engine after the allocation of stream data and CEP rules. In this paper, we suggest an efficient rule allocation algorithm to prevent such situations. This algorithm determines event rules priorities for the allocation, wherein the rule with higher priority is allocated first to the engine that minimizes the increase of the value of the proposed cost function. We prove the superiority of our algorithm in two tests. In the optimization verification test, our algorithm achieves the results closest to the optimal results compared with the other algorithms. In the performance test, our algorithm shows lower latency and data replication ratio in the distributed CEP system using real world dataset and event rules.
Parallel Computation of Order-Preserving Periods and Order-Preserving Borders of a Set of Strings
http://doi.org/10.5626/JOK.2019.46.12.1232
Given two strings of the same length over an integer alphabet, those two strings are order-isomorphic when they have the same relative ranks. When strings order-isomorphic to T[1..p](1 ≤ p ≤ n) are periodically repeated in T, a representation of the order relations of T[1..p] is referred to as an order-preserving period of T. When a prefix T[1..q] (1 ≤ q ≤ n) of T is order-isomorphic to a suffix T[n-q+1..n] of T, a representation of the order relations of T[1..q] is called an order-preserving border of T. The lengths of all order-preserving periods (resp. all order-preserving borders) of T can be computed in O(n logn) time using the Z-function. Given a set Ŝ=｛S₁, S₂,..., Sr｝of strings of length n over an integer alphabet, we propose parallel algorithms computing the lengths of all order-preserving periods and all order-preserving borders of Ŝ using O(rn) threads in O(n) time by the Z-function. When compared to each sequential algorithm for Dow Jones Industrial Average, the experimental results show that our parallel algorithm for computing the lengths of all order-preserving periods (resp. all order-preserving borders) of Ŝ runs approximately 3.47 (resp. 3.41) times faster when r =1,000, n =10,000.
Efficient CNNs with Channel Attention and Group Convolution for Facial Expression Recognition
MyeongOh Lee, Ui Nyoung Yoon, Seunghyun Ko, Geun-Sik Jo
http://doi.org/10.5626/JOK.2019.46.12.1241
Recently, studies using the convolutional neural network have been actively conducted to recognize emotions from facial expressions. In this paper, we propose an efficient convolutional neural network that solves the model complexity problem of the deep convolutional neural network used to recognize the emotions in facial expression. To reduce the complexity of the model, we used group convolution, depth-wise separable convolution to reduce the number of parameters, and the computational cost. We also enhanced the reuse of features and channel information by using Skip Connection for feature connection and Channel Attention. Our method achieved 70.32% and 85.23% accuracy on FER2013, RAF-single datasets with four times fewer parameters (0.39 Million, 0.41 Million) than the existing model.
Forecast of the Stock Market Price using Artificial Neural Network and Wavelet Transform
http://doi.org/10.5626/JOK.2019.46.12.1249
With advancements in technologies on machine learning and artificial neural network, various researches have attempted to predict the changes in the price of the stock market. The prediction accuracy has improved with adoption of new artificial neural network technologies that have been developed for image and voice signal processing. In the present work, the technical indices from KOSPI were decomposed for the prediction of index and movement direction of KOSPI into high-frequency part and low-frequency part using wavelet transform, then used to predict KOSPI independently by using artificial neural networks. For the final prediction, the prediction result of each frequency part was added. CNN, DPN, and LSTM were employed as artificial neural network; the performance of each model was compared and the efficiency of the wavelet transform of input variables was analyzed. CNN with 0.51% of MAPE for the index prediction and LSTM with 81.7% of accuracy for movement prediction showed the best performance among the three models. The efficiency of wavelet transform was confirmed with averaged 38% of the improved performance for the index prediction and averaged 25% of the improved performance for the movement prediction.
Integrated Hazard Analysis Process for Safety and Security based on SysML
http://doi.org/10.5626/JOK.2019.46.12.1262
In a safety-critical system, an accident may cause harm to people and the environment. Therefore, it is important to thoroughly analyze potential hazards and elicit safety requirements from the concept phase of the system to be developed. Traditionally, component failure has been considered a major hazard. However, in modern systems, software faults and failed interactions among components are other major hazards that must be considered. As wired/wireless network connection plays a major role in recent systems, cyber security has become a major system safety concern. Such threats from hackers should be considered in hazard analysis. In the past, hazard analysis in safety and threat analysis in security have been treated as independent activities. As reports about the common assets and the complementary features of both techniques have been released, combining these two analysis techniques has attracted interest in the safety area. The major focuses of this study involved the analysis techniques and the assessment methodologies. Prior studies that have combined hazard and threat analyses have not provide systematic processes that can be followed by practitioners, which is a critical inconvenience in developing safety critical systems for the field. In this paper, we propose a hazard analysis process based on SAMM that integrates threats related to safety using SysML diagrams. We applied the proposed process to a remote parking assistance system to evaluate its effectiveness.
Developing a Connection Restrictions Filtering System for Websites based on Swear Words Extraction
http://doi.org/10.5626/JOK.2019.46.12.1272
Youth are exposed to various types of illegal and harmful information through the Internet. To reduce exposure to such information, the government adopts the SNI method to block access to illegal harmful sites. However, it does not block instances of bad information or swear words on the web page itself. In order to limit the access of these web pages in situations such as schools and institutions, this study suggests a connection restrictions filtering system for websites based on swear words extraction. We collected 5542 pseudonyms that were actually used in related research, questionnaires, and Internet searches. We extracted the profanity by using the w-shingling algorithm, then calculated the risk associated with the webpage according to the frequency of use and the weight of the profanity. The system developed in this study will help learning environments in small networks such as elementary and junior high schools by allowing them to restrict access to websites for educational purposes.
Lightweight Equivalence Checking of Code Transformation for Code Pointer Integrity
Jaeseo Lee, Tae-Hyoung Choi, Gyuho Lee, Jaegwan Yu, Kyungmin Bae
http://doi.org/10.5626/JOK.2019.46.12.1279
Code transformation is widely used to improve the performance and security of programs, but serious software errors can occur in this process if the generated program is not equivalent to the original program. There are a number of approaches for translation validation that can be used to prove the equivalence of programs, but the high cost of proof checking restricts the applicability of these techniques for large programs. In this paper, we propose a practical approach for checking the correctness of LLVM code transformation. We first prove the correctness of the transformation rules using automated theorem proving before compilation. We then perform a simple code analysis method—as opposed to directly proving the program equivalence— to check whether the transformations rules are correctly applied to the generated code. As the complexity of the proposed code analysis is linear, our technique can be effectively applied to large programs, unlike previous techniques. To prove the effectiveness of the proposed approach, we present a case study on LLVM code transformation for a code pointer integrity instrumentation.
A DNN-based Epileptic EEG Detection System for Epileptic Patient Classification
Won Jun Park, Jin Hyeok Park, Young Ho Lee
http://doi.org/10.5626/JOK.2019.46.12.1291
Epilepsy is a brain disease that causes the whole body to twist with bubbles around the mouth. Patient with epilepsy can lose consciousness. About 40 to 50 million people around the world suffer from epilepsy. Because epilepsy is unpredictable, epileptic patients are always exposed to the risk of physical damage. Therefore, it is important to classify epileptic patients before they have seizures to prevent accidents. In this study, we propose a DNN - based epileptic EEG detection system to classify epilepsy patients. EEG data from normal and epileptic patients were used for this study. In data preprocessing, the ADASYN technique was used to reduce data imbalance and important features were extracted using the IPCA technique. These extracted features were applied to four optimization algorithms of deep learning. A model was constructed and its performance was evaluated. Experimental results showed that the optimized model using the Nadam algorithm had the highest performance with an accuracy of 97.6% and an AUC value of 0.997. Using this optimized model, physicians will be able to diagnose epilepsy patients with high accuracy during EEG.
Wave Celerity Estimation using Unsupervised Image Registration from Video Imagery
Jinah Kim, Jaeil Kim, Sungwon Shin
http://doi.org/10.5626/JOK.2019.46.12.1296
In this paper, we propose an image registration method based on unsupervised learning to estimate wave celerity by tracking wave movements using a large amount of video imagery. It is difficult to estimate the wave celerity accurately using physics-based modeling in the coastal region, owing to the limitations of in-situ measurement and the high nonlinearity of wave phenomena itself as well as high complexity from nonlinear interactions. In order to estimate wave celerity, the proposed method learns the nonlinear wave behavior from the video imagery. Autoencoder is applied to separate hydrodynamics scenes from environmental factors, such as daylights. The displacement vector of propagating waves is computed by non-linear spatio-temporal image registration. The wave celerity is estimated by accumulating the displacement vectors along time. In this paper, we compare the wave celerity measurement with conventional image processing methods and actual measurement using sensors for accuracy evaluation.
Neural Module Network Learning for Visual Dialog
http://doi.org/10.5626/JOK.2019.46.12.1304
In this paper, we propose a novel neural module network (NMN) model for visual dialog. Visual dialog currently has several challenges: The first one is visual grounding, which is concerned with how to associate the entities mentioned in the natural language question with the visual objects included in the given image. The other one is visual co-reference resolution, which involves how to determine which words, typically noun phrases and pronouns, co-refer to the same visual object in a given image. In order to address these issues, we suggest a new visual dialog model using both question-customized neural module networks and a reference pool. The proposed model includes not only a new Compare module to answer the questions that require comparing prosperities between two visual objects, but also a novel Find module improved by using a dual attention mechanism, and a Refer module to resolve visual co-references with the reference pool. To evaluate the performance of the proposed model, we conduct various experiments on two large benchmark datasets, VisDial v0.9 and VisDial v1.0. The results of these experiments show that the proposed model outperforms the state-of-the-art models for visual dialog.
Document Image Binarization Using Multi-scale Fusion Network
http://doi.org/10.5626/JOK.2019.46.12.1314
Binarization of degraded document images has a significant impact on document image analysis domains. We developed a LadderNet with multi-scale architecture to learn features from degraded document images to classify text and background from noise pixels. Then, binarized images are generated from this classification. Specifically, we consider two properly designed LadderNet architectures: One with deeper architecture, another with shallower architecture. Each structure is calibrated independently using document image patches. A predicted output generated from the deeper architecture with smaller size of the striding window has clearer text-strokes but contains marked noise. Conversely, a predicted output has lower noise in the background from the shallower architecture with larger size of the striding window. However, the detail of the text is not clear. Thus, a better result is achieved by combining the outputs of two of these architectures. We tested the proposed model on benchmark DIBCO datasets for document image binarization and achieved superior performance over existing methods in the literature.
Optimization of Distributed Binary Bernoulli Sampling
Wonhyeong Cho, Myeong-Seon Gil, Namsu Ju, Yang-Sae Moon
http://doi.org/10.5626/JOK.2019.46.12.1322
This paper proposes a method to improve the performance of Binary Bernoulli Sampling(BBS). BBS is a sampling technique suitable for a multi-source stream environment. Accordingly, a recent approach has been proposed for distributed processing of BBS based on Apache Storm, with a multi-coordinator structure. However, this approach causes an additional coordinator waiting problem, which limits the performance improvement. In this paper, we solve the coordinator waiting problem by introducing a multi-distribution structure and a distributor separation structure. The multi-distribution structure enables multiple coordinators, rather than one, to participate in the distribution, minimizing the coordinator waiting time. The distributor separation structure moves the distributing function from the coordinators to the distributors, maximizing the processing performance. We perform various experiments by implementing our proposed structure on the Storm-based distributed BBS. The experimental results show that our structure improves the performance by up to 90 times compared to the previous distributed BBS.
Analysis and Modeling of Advanced Persistent Threat through Case Study
http://doi.org/10.5626/JOK.2019.46.12.1328
Advanced Persistent Threat(APT) attack is one of the cyber-attack methods that continuously attacks the specific target with advanced tools. Since attackers use various methods that are specialized to targets, it is difficult to prevent the attacks with common security countermeasures. Currently, there exist various the APT attack stage models. However, the models only express APT attacks simply. Consequently, it is difficult to use them for risk assessment or as a recommendation for security requirements for a specific system. In order to overcome the limitations of such models, we derived factors of APT attack through a case study for defining the features of APT attack. We have also analyzed and defined the factors and their relationships to construct the APT attack factor model. For validation purpose, the model applied to the actual attack case has been referred to as ‘APT 1’. Through the proposed model, it would be possible to gain understanding about the overall flow of APT attacks and classify attack factors not only in terms of technical aspects but also with respect to social engineering facets.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr