• KIPS Transactions on Computer and Communication Systems
• /
• v.11 no.2
• /
• pp.43-50
• /
• 2022

As the use of mobile data increases, the proportion of video content is increasing steeply. In order to solve problems that arise when mobile users receive data from geographically remote cloud servers, methods of caching data in advance to edge servers geographically close to the users are attracting lots of attention. In this paper, we present a caching policy that stores data on Small Cell Base Station(SBS) to effectively provide content files to mobile users by applying a delayed offloading scheme in a cellular network. The goal of the proposed policy is to minimize the size of data transmitted from Macro Base Station(MBS) because the delayed offloading scheme requires more cost than when downloaded from MBS than from SBS. The caching policy is proposed to determine the size of content file and which content file to be cached to SBS using the probability of mobile users' paths and the popularity of content files, and to replace content files in consideration of the overlapping coverage of SBS. In addition, through performance evaluation, it has been proven that the proposed policy reduces the size of data downloaded from MBS compared to other algorithms.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.10
• /
• pp.67-76
• /
• 2023

Although deep learning models are making innovative achievements in the field of computer vision, the problem of vulnerability to adversarial examples continues to be raised. Adversarial examples are attack methods that inject fine noise into images to induce misclassification, which can pose a serious threat to the application of deep learning models in the real world. In this paper, we propose a model that detects adversarial examples using differences in predictive values between edge-learned classification models and underlying classification models. The simple process of extracting the edges of the objects and reflecting them in learning can increase the robustness of the classification model, and economical and efficient detection is possible by detecting adversarial examples through differences in predictions between models. In our experiments, the general model showed accuracy of {49.9%, 29.84%, 18.46%, 4.95%, 3.36%} for adversarial examples (eps={0.02, 0.05, 0.1, 0.2, 0.3}), whereas the Canny edge model showed accuracy of {82.58%, 65.96%, 46.71%, 24.94%, 13.41%} and other edge models showed a similar level of accuracy also, indicating that the edge model was more robust against adversarial examples. In addition, adversarial example detection using differences in predictions between models revealed detection rates of {85.47%, 84.64%, 91.44%, 95.47%, and 87.61%} for each epsilon-specific adversarial example. It is expected that this study will contribute to improving the reliability of deep learning models in related research and application industries such as medical, autonomous driving, security, and national defense.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2012.07a
• /
• pp.71-74
• /
• 2012

본 논문에서는 새로운 문자 분할 알고리즘을 제안한다. 고전적인 문자 분할 알고리즘은 학술적인 문서영상과 같이 단순한 구조를 가진 문서영상을 대상으로 하여 좋은 성능을 보였지만 다양한 문자 크기와 색상, 그림, 복잡한 배경 등으로 구성된 문서영상에서는 좋지 못한 성능을 보인다. 최근에 제안고 있는 방법들은 복잡한 문서영상에서도 좋은 성능을 보이도록 다양한 기법들을 적용하여 우수한 성능을 보이고 있지만, 대부분의 방법들이 영상을 일정한 크기의 블록으로 나누어 문자분할을 하기 때문에 세밀한 부분에서는 성능이 어느 정도 한계를 보인다. 따라서 본 논문에서는 블록의 크기에 제한을 갖지 않는 새로운 방법으로서, watershed 알고리즘을 이용한 문자분할 방법을 제시한다. 구체적으로, watershed 알고리즘을 이용하여 문서영상의 구조(docstrum)를 파악하고 이를 기반으로 문자를 분할한다. 제안하는 방법은 크게 엣지 검출, distance transform, watershed 알고리즘을 이용한 docstrum 분석, 문자 분할의 네 단계를 거친다. 실험 결과 블록에 기반한 기존의 방법들이 놓치는 세밀한 부분에서도 제안된 알고리즘은 올바른 분할결과를 얻을 수 있음을 확인하였다.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.427-432
• /
• 2020

딥러닝(Deep-learning) 기반의 자연어 이해(Natural Language Understanding) 기술들은 최근에 상당한 성과를 성취했다. 하지만 딥러닝 기반의 자연어 이해 기술들은 내적인 동작들과 결정에 대한 근거를 설명하기 어렵다. 본 논문에서는 벡터를 그래프로 변환함으로써 신경망의 내적인 의미 표현들을 설명할 수 있도록 한다. 먼저 인간과 기계 모두가 이해 가능한 표현방법의 하나로 그래프를 주요 표현방법으로 선택하였다. 또한 그래프의 구성요소인 노드(Node) 및 엣지(Edge)의 결정을 위한 Element-Importance Inverse-Semantic-Importance(EI-ISI) 점수와 Element-Element-Correlation(EEC) 점수를 심층신경망의 훈련방법 중 하나인 드랍아웃(Dropout)을 통해 계산하는 방법을 제안한다. 다양한 실험들을 통해, 본 연구에서 제안한 벡터-그래프(Vector2graph) 변환 프레임워크가 성공적으로 벡터의 의미정보를 유지하면서도, 설명 가능한 그래프를 생성함을 보인다. 더불어, 그래프 기반의 새로운 시각화 방법을 소개한다.

• Journal of radiological science and technology
• /
• v.43 no.4
• /
• pp.251-257
• /
• 2020

This study was purpose to quantitative evaluation of edge method of modulation transfer function(MTF) and physical image characteristics of by obtain the optimal edge image by using magnetic resonance imaging(MRI). The MRI equipment was used (MAGNETOM Vida 3.0T MRI, Siemense healthcare system, Germany) and the head/neck matrix shim MR coil were 20 channels(elements) receive coil. The MTF results of showed the best value of 0.294 based on the T2 Nyquist frequency of 1.0 mm^-1. The MTF results of showed that the T1 image is 0.160, the T1 CE image is 0.250, T1 Conca2 image is 0.043, and the T1 CE (Concatenation) Conca2 image is 0.190. The T2 image highest quantitatively value for MTF. The physical image characteristics of this study were to that can be used efficiently of the MRI and to present the quantitative evaluation method and physical image characteristics of 3.0T MRI.

• KIPS Transactions on Software and Data Engineering
• /
• v.2 no.3
• /
• pp.195-198
• /
• 2013

Many libraries utilize RFID tags for checking in and out of books. However, the recognition rate of this automatic process may depend on the orientation of antennas and RFID tags. Therefore we need supplemental systems to improve the recognition rate. The proposed algorithm sets up the ROI of the book existing area from the input image and then performs Canny edge detection algorithm to extract edges of books. Finally Hough line transform algorithm allows to detect the number of books from the extracted edges. To evaluate the performance of the proposed method, we applied our method to 350 book images under various circumstances. We then analyzed the performance of proposed method from results using recognition and mismatch ratio. The experimental result gave us 97.1% accuracy in book counting.

• Journal of radiological science and technology
• /
• v.42 no.5
• /
• pp.329-334
• /
• 2019

This study was purpose to quantitative evaluation of noise power spectrum(NPS) and studied the quantitative evaluation and characteristics of modulation transfer function(MTF) by obtain the optimal edge image by using Coil in magnetic resonance imaging(MRI) equipment through Fujita theory using edge method. The MRI equipment was used (Tim AVANTO 1.5T, Siemense healthcare system, Germany) and the head matrix coil were 12channels(elements) receive coil. The NPS results of showed the best value of 0.004 based on the T2 Nyquist frequency of $1.0mm^{-1}$, and the MTF results of showed that the T1 and T2 values were generally better than the T1 CE and T1 CE FC values. The characteristics of this study were to explain the characteristic method of image quality evaluation in general. To present the quantitative evaluation process and results in the evaluation of MRI image characteristics in radiology.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.10
• /
• pp.1234-1239
• /
• 2019

This paper proposes a method for measuring fine dust concentration through digital processing of images captured by only existing CCTVs without additional equipment. This image processing algorithm consists of noise reduction, edge sharpening, ROI setting, edge strength calculation, and correction through HSV conversion. This algorithm is implemented using the C ++ OpenCV library. The algorithm was applied to CCTV images captured over a month. The edge strength values calculated for the ROI region are found to be closely related to the fine dust concentration data. To infer the correlation between the two types fo data, a trend line in the form of a power equation is established using MATLAB. The number of data points deviating from the trend line accounts for around 12.5%. Therefore, the overall accuracy is about 87.5%.

• Journal of radiological science and technology
• /
• v.45 no.1
• /
• pp.11-17
• /
• 2022

This study was purpose to quantitative assessment of the resolution characteristics by using American college of radiology(ACR) phantom for magnetic resonance imaging (MRI). The MRI equipment was used (Achiva 3.0T MRI, Philips system, Netherlands) and the head/neck matrix shim SENSE head coil were 32 channels(elements) receive MR coil. And the MRI equipment was used (Discovery MR 750, 3.0T MRI, GE medical system, America) and the head/neck matrix shim MC 3003G-32R 32-CH head coil were receive MR coil. As for the modulation transfer function(MTF) comparison result by using ACR magnetic resonance imaging phantom, the MTF value of the ACR standard T2 image in GE equipment is 0.199 when the frequency is 1.0 mm^-1 and the MTF value of the hospital T2 image in Philips equipment is 0.528. It was used efficiently by using a general sequence more than the standard sequence method using the ACR phantom. In addition it is significant that the quantitative quality assurance evaluation method for resolution characteristics was applied mutatis mutandis, and the result values of the physical image characteristics of the 3.0T MRI device were presented.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.18 no.6
• /
• pp.728-735
• /
• 2017

With the proliferation of digital devices, the devices have generated considerable additive white Gaussian noise while acquiring digital images. The most well-known denoising methods focused on eliminating the noise, so detailed components that include image information were removed proportionally while eliminating the image noise. The proposed algorithm provides a method that preserves the details and effectively removes the noise. In this proposed method, the goal is to separate meaningful detail information in image noise environment using the edge strength and edge connectivity. Consequently, even as the noise level increases, it shows denoising results better than the other benchmark methods because proposed method extracts the connected detail component information. In addition, the proposed method effectively eliminated the noise for various noise levels; compared to the benchmark algorithms, the proposed algorithm shows a highly structural similarity index(SSIM) value and peak signal-to-noise ratio(PSNR) value, respectively. As shown the result of high SSIMs, it was confirmed that the SSIMs of the denoising results includes a human visual system(HVS).