http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
이수현(Soo-Hyeon Lee),김준수(Junsu Kim),황현욱(Hyunuk Hwang),이해연(Hae-Yeoun Lee) 한국디지털포렌식학회 2021 디지털 포렌식 연구 Vol.15 No.1
IT 기술의 발전으로 많은 이미지 데이터들이 생산 및 유통되고 있어서, 디지털포렌식에서 처리해야 하는 데이터의 양이 급증하고 있어서 활용할 수 있는 인적 및 물적 자원의 한계에 도달하고 있다. 문서 스캔 이미지는 포렌식 분석의 주요한 대상으로서, 방대한 양의 이미지 데이터에서 문서 스캔 이미지들만을 탐색하는 것은 중요하지만, 범용적인 딥 러닝 기술의 경우 포렌식에서 요구하는 정확도를 충족하지 못하는 문제가 있다. 본 논문에서는 딥 러닝을 이용하여 문서 스캔 이미지를 탐색하는 기술에 대하여 설명하고, 실제 수집된 문서 스캔 이미지를 대상으로 탐색한 결과를 제시한다. 영상 인식과 분류에서 범용적으로 사용되는 컨볼루셔널 뉴럴 네트워크 기반의 딥 러닝 모델을 도입하였고, 문서 스캔 이미지 탐색에 적합하도록 컨볼루셔널 및 전연결 계층에 대한 설계를 하였으며, 활성화 함수 및 풀링 함수 등의 설정을 하였다. MSCOCO 데이터셋에서 가져온 일반 이미지들과 5종으로 정의하여 수집한 문서 스캔 이미지를 이용하여 데이터베이스를 구축하였고, 이를 바탕으로 제시한 딥 러닝 모델에 대하여 학습을 수행한 후에 탐색 성능에 대한 분석을 수행하였고 평균적으로 98.89%의 탐색 정확도를 달성하였다. As many images are created and distributed with advances in IT technology, the amount of data that must be processed with digital forensics is rapidly increasing and available human and material resources are reaching their limits. Document scan images are a major target of forensic analysis and it is important to search these images from a vast amount of images. However, general-purpose deep learning techniques do not meet the accuracy required for forensics. In this paper, we describe a document scan image searching technique using deep learning, and present the results of searching the actually collected document scan images as targets. A convolutional neural network-based deep learning model, which is commonly used in image recognition and classification, was adapted. To be suitable for document scan image searching, convolutional and fully-connected layers were designed and activation functions and pooling functions, etc. were set. A database was constructed using general images of MSCOCO dataset and document scan images collected as 5 types. After training the presented deep learning model using this database, the searching performance was analyzed and the average searching accuracy of 98.89% was achieved.
이수현(Soo-Hyeon Lee),이해연(Hae-Yeoun Lee) 대한전자공학회 2019 대한전자공학회 학술대회 Vol.2019 No.11
Camera model discrimination technology has occupied a major position in the digital forensic technique. Research has been conducted in a variety of ways, most with over 95% accuracy. In this paper, we focused on the fact that the subject is in the center of the image and the edges of the image contain contents that are difficult to find feature of image, such as mountains, sky and sea. Therefore, performance analysis was performed for both the central area of image and all area of image. As a result, it was confirmed that using the center portion of the image shows a higher performance.
고기능 자폐 범주성 장애 아동의 SNS 대화 맥락에서 유생성에 따른 이모티콘 표현 능력 및 선호도 분석
이수현(Soo Hyeon Lee),김영태(Young Tae Kim),연석정(Seok Jeong Yeon) 한국언어치료학회 2017 言語治療硏究 Vol.26 No.2
Purpose: The purpose of this study was to investigate the expression and preference of emoticons according to the animacy between children with high-functioning autism spectrum disorder (HFA) and typically developing children (TD) in SNS contexts. Methods: Twenty-four children (12 HFA, 12 TD) aged from 6 to 10 years participated in the study. The emoticon expression task consisted of facial and object emoticon tasks. SNS conversations of two communicators were provided to children and they were asked to choose one answer from four selection type emotions on screen. Two-way mixed analysis of variance (ANOVA) was used to investigate the group differences in the emoticon expression ability according to the animacy in SNS conversations. Results: First, children with HFA showed a significantly lower performance of emoticon expression task than TD children. This finding explains why children with HFA have poorer expression ability of emoticons than those of TD children. However, there was no significant difference in the emoticon type (facial vs. object) according to animacy. Second, children with HFA showed a significantly lower preference on facial emoticons than TD children. Conclusions: These results imply that children with HFA have a weak expression ability of emoticons in SNS contexts. However, in facial emoticon expression tasks, HFA showed significantly higher performance than tasks with object emoticons. In preference tasks of facial emoticons in SNS contexts, HFA showed a significantly lower preference percentage than TD. Therefore, the characteristics of HFA using emoticons in SNS contexts is considered for social communication intervention.
증례 : 위상피 종양의 내시경 점막절제술 후 치유되지 않은 인공 위궤양
이수현 ( Soo Hyeon Lee ),천재희 ( Jae Hee Cheon ),김지현 ( Jie Hyun Kim ),박종필 ( Jong Pill Park ),이상길 ( Sang Kil Lee ),이용찬 ( Yong Chan Lee ) 대한소화기학회 2007 대한소화기학회지 Vol.51 No.2
Endoscopic mucosal resection (EMR) is widely accepted as a standard treatment for early gastric cancer or gastric adenoma. However, EMR inevitably results in the formation of large iatrogenic ulcer at the resected area. Although the characteristics of EMR-induced ulceration are not fully understood, this type of ulcer is thought to heal faster and to recur less often than non-iatrogenic gastric ulcer. Current available evidences have suggested that EMR-induced ulcers heal within 2-3 months. Herein, we report two cases of non-healing persistent gastric ulcers after EMR. One is a case of gastric carcinoma which developed at the same site of previous EMR site for the low grade dysplasia. The other is a case in which persistent EMR-induced ulcer was healed in the long run after Helicobacter pylori eradication therapy. (Korean J Gastroenterol 2008;51:127-131)
신규현(Shin Kyu Hyeon),이용연(Lee Yong Yeun),이수한(Lee Soo Han) 대한기계학회 2004 대한기계학회 춘추학술대회 Vol.2004 No.11
In this paper, a family of decentralized adaptive controller is proposed to control robot manipulators which are governed by highly nonlinear dynamic equations. The controller is computationally efficient since it does not require mathematical model or parameter values of the manipulators. The stability of the manipulators with the controller is proved by Lyapunov theory. The results of numerical simulations show that the system is stable, and has excellent trajectory tracking performance.