http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
PNG 영상에서 JPEG 영상으로의 변환에 대한 압축률에 따른 품질 분석
시종욱(Jongwook Si),김성영(Sungyoung Kim) 한국정보기술학회 2021 Proceedings of KIIT Conference Vol.2021 No.11
영상 파일 형식은 디지털 정보를 저장하는 방법으로 다양한 종류의 형식이 존재한다. 그런데 다양한 파일형식으로 인해 변환기를 사용해 포맷을 변경하고 용량 제한이 있는 경우 압축이 필요하다. 하지만 높은 압축률로 변환하는 경우 본래의 이미지의 화질이 저하되어 사용하기 힘들다. 따라서 본 논문에서는 PNG 영상에서 JPEG 영상으로 압축률에 따른 변환에 대하여 영상 품질 지표를 이용해 분석한 결과를 제시한다. The image file format is a method of storing digital information, and various types of formats exist. However, due to various file types, compression is required if the format is changed using a converter and there is a capacity limit. However, when converting to a high compression rate, the image quality of the original image is deteriorated and it is difficult to use. Therefore, in this paper, the results of analysis using image quality indicators for conversion according to the compression rate from PNG image to JPEG image are presented.
통계 기반 특징 선택과 적대적 생성 학습을 이용한 사출기의 이상 감지
시종욱(Jongwook Si),정지수(Jisu Jeong),정민수(Minsu Jeong),김성영(Sungyoung Kim) 한국정보기술학회 2023 한국정보기술학회논문지 Vol.21 No.4
The automation system is being built in the manufacturing industry, and anomaly detection in defective products should be included. Among the manufacturing industry, mold manufacturing goes through a process of producing a large amount of products through injection molding. So, it is possible to detect the production of defective products for the mold through injection molding. In this paper, we introduce a method for anomaly detection using the data of sensors attached to the injection molding. In data feature selection, only meaningful features are selected based on a statistic, and anomaly detection models through generative adversarial learning are trained and inferred. Data were collected every minute through sensors attached to injection molding, and existing models were tested and analyzed. As a result, it was confirmed that the feature selection can improve performance of the injection molding.
골격 좌표 벡터 및 LSTM 모델을 이용한 넘어짐 검출
시종욱(Jongwook Si),손현철(Hyeoncheol Son),김다슬(Daseul Kim),김문년(Moonnyeon Kim),정지연(Jiyeon Jeong),김규리(Gyuree Kim),김영형(Younghyung Kim),김성영(Sungyoung Kim) 한국정보기술학회 2020 한국정보기술학회논문지 Vol.18 No.12
In recent years, many countries around the world have faced the problems of aging, single-person households, and the increasing number of elderly people who live alone. These problems have led to the need for research to improve the quality of life for the elder and single-person households. In this paper, the method of detecting falls was proposed with a focus on the aspect of safety among the methods of improving the quality of life of single-person households. First, we extract key points of human skeletons based on the existing method, vectorize them to represent correlation between them, and input them into LSTM to determine whether falls have occurred. We try to enhance the performance by using only representative feature points, not all feature points. Five reference datasets were used to evaluate the performance of proposed system and performed well in most datasets.
단일 입출력 심볼로 구성된 순환형 전자 회로도 영상의 네트리스트 변환
시종욱(Jongwook Si),김문년(Moonnyeon Kim),김성영(Sungyoung Kim) 한국정보기술학회 2021 한국정보기술학회논문지 Vol.19 No.8
Electronic drawings and related books contain many circuit diagrams and require circuit configuration to see the results. However, it uses a lot of computer simulation programs because it is a waste of money and labor that is actually constructed. However, simulation programs require passive tasks such as placing circuits and entering values by the user. Therefore, in this paper, we have developed a new method to convert drawing images of electronic circuits into PSPICE Netlists. First, circuit symbols are recognized through learning about circuit symbols. We then use image processing to analyze the relationship between wires connecting the recognized circuit symbols based on the drawing where the symbols were removed. Finally, we construct Netlists by combining the symbol recognition and wire analysis information. The results of the proposed method show high performance on cyclic electronic circuit diagrams consisting of a single I/O.