• 정보처리학회논문지:소프트웨어 및 데이터공학
• /
• 제10권11호
• /
• pp.473-482
• /
• 2021

최근 딥러닝을 이용한 한글 생성 모델이 연구되고 있으나, 한글 폰트의 구조가 복잡하고 많은 폰트 데이터가 필요하여 상당한 시간과 자원을 필요로 할 뿐 아니라 스타일이 제대로 변환되지 않는 경우도 발생한다. 이러한 문제점을 보완하기 위하여, 본 논문에서는 한글의 초성, 중성, 종성의 구성요소를 기반으로 최소 글자를 사용하는 한글 폰트 생성 모델인 CKFont 모델을 제안한다. CKFont 모델은 GAN을 사용하는 한글 자동 생성 모델로, 28개의 글자와 초/중/종성 구성요소를 이용하여 다양한 스타일의 모든 한글을 생성할 수 있다. 구성요소로부터 로컬 스타일 정보를 획득함으로써, 글로벌 정보 획득보다 정확하고 정보 손실을 줄일 수 있다. 실험 결과 스타일을 자연스럽게 변환되지 못하는 경우를 감소시키고 폰트의 품질이 향상되었다. 한글 폰트를 생성하는 다른 모델들과 비교하여, 본 연구에서 제안하는 CKFont는 최소 글자를 사용하는 모델로, 모델의 구조가 간결하여 폰트를 생성하는 시간과 자원이 절약되는 효율적인 모델이다. 구성요소를 이용하는 방법은 다른 언어 폰트의 변환은 물론 다양한 이미지 변환과 합성에도 사용될 수 있다.

• Journal of Positioning, Navigation, and Timing
• /
• 제11권3호
• /
• pp.163-172
• /
• 2022

Typical anti-jamming technologies based on array antennas, Space Time Adaptive Process (STAP) & Space Frequency Adaptive Process (SFAP), are very effective algorithms to perform nulling and beamforming. However, it does not perform equally well for all types of jamming signals. If the anti-jamming algorithm is not optimized for each signal type, anti-jamming performance deteriorates and the operation stability of the system become worse by unnecessary computation. Therefore, jamming classification technique is required to obtain optimal anti-jamming performance. Machine learning, which has recently been in the spotlight, can be considered to classify jamming signal. In general, performing supervised learning for classification requires a huge amount of data and new learning for unfamiliar signal. In the case of jamming signal classification, it is difficult to obtain large amount of data because outdoor jamming signal reception environment is difficult to configure and the signal type of attacker is unknown. Therefore, this paper proposes few-shot jamming signal classification technique using meta-learning and transfer-learning to train the model using a small amount of data. A training dataset is constructed by anti-jamming algorithm input data within the GNSS receiver when jamming signals are applied. For meta-learning, Model-Agnostic Meta-Learning (MAML) algorithm with a general Convolution Neural Networks (CNN) model is used, and the same CNN model is used for transfer-learning. They are trained through episodic training using training datasets on developed our Python-based simulator. The results show both algorithms can be trained with less data and immediately respond to new signal types. Also, the performances of two algorithms are compared to determine which algorithm is more suitable for classifying jamming signals.

• 한국콘텐츠학회논문지
• /
• 제9권10호
• /
• pp.159-167
• /
• 2009

방송에서 TV 드라마의 비중과 영향력이 커지면서, TV 드라마에 대한 많은 연구들이 이루어지기 시작했다. 그러나 이들 연구들은 시청률에 영향을 주는 시청동기, 수용행태 등에 관한 수용자 연구와 드라마 서사구조의 사회적 의미를 분석한 것들이 대부분이었고, TV 드라마의 영상제작기법과 연관된 연구는 시청률과 영상제작기법의 상관성, 시대별 영상제작기법 변천과정, 영화와 TV드라마의 영상제작기법 비교 등과 같은 제한된 수의 연구들이 이루어졌지만, 드라마의 내용과 영상제작기법의 관계를 조명한 연구는 전문한 실정이다. 따라서 본 연구에서는 텔레비전 편성에서 뿐만 아니라 산업적 가치 측면에서도 그 비중이 커지고 있는 텔레비전 드라마의 내용상의 장르와 영상제작기법의 상관성을 분석해보고자 한다. 표본 추출을 위하여 TNS 미디어 코리아에서 제시한 2004년부터 2008년까지의 연간 '시청률 톱 100' 자료를 바탕으로, 이 기간 동안 방송된 역사드라마, 멜로드라마, 홈드라마 중 가장 시청률이 높은 드라마를 각 장르별로 5편씩 총 15편의 드라마에서 8,210개 샷들을 추출하여 이들의 영상제작기법을 비교 분석하였다. 분석결과 역사드라마에서는 홈드라마나 멜로드라마에 비해 클로즈업과 롱샷 그리고 트래킹을 상대적으로 많이 사용하였으며, 샷의 지속시간은 짧은 반면 장면의 지속시간은 긴 것으로 나타났다. 반면 홈 드라마와 멜로드라마에서는 역사드라마에 비해 웨이스트샷이 상대적으로 많이 사용되었으며, 샷의 지속시간은 긴반면, 장면의 지속시간은 상대적으로 짧게 나타났다.

• 한국멀티미디어학회논문지
• /
• 제23권1호
• /
• pp.43-49
• /
• 2020

Along side with the rapid growth of local governments' advertising bills, there are few studies focused on the effectiveness of these ads. Especially one of the media being used by the local governments is the Korea Express Train (KTX), where they advertise in the train coaches' KTX video monitor. Unfortunately the ads in KTX are exposed without audio mostly. The current study, therefore, probed on the effectiveness of these ads. This study utilized transportation theory and content analysis methodology to give insight to its discourse. We established two analysis units (camera and subtitles), and then analyzed 107 local government ads. From the camera analysis, it is observed that local governments' festival and tour promotion ads more often employ dynamic angles such as drone shot and long shot. Also, from subtitles usage analysis, it is observed that many of the ads make use of large size titles and subtitles which could prevent viewers seeing visual shots. In the special case audio-less KTX ads, this study recommends emphasis on subtitles which will enhance the ad effectiveness of the ad messages.

• 문화기술의 융합
• /
• 제8권1호
• /
• pp.545-550
• /
• 2022

딥러닝은 많은 데이터를 필요로 한다는 것은 이미 널리 알려져있다. 이를 통해, 딥러닝에 쓰이는 신경망의 수없이 많은 parameter들을 학습시킨다. 학습과정에는 데이터뿐 아니라, 각 데이터별로 전문가가 입력한 label이 필요한 경우가 대부분인데, 이 label을 얻는 과정은 시간과 자원 소비가 심하다. 이 문제를 완화하기 위해, few-shot learning, self-supervised learning, weak-supervised learning등이 연구되어오고 있다. 본 논문에서는, label을 상대적으로 적은 노력으로 수행하기 위한 연구들의 동향을 살펴보고, 앞으로의 개선 방향을 제시하도록 한다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권6호
• /
• pp.3074-3091
• /
• 2019

With the widespread use of the Internet, services for providing large-capacity multimedia data such as video-on-demand (VOD) services and video uploading sites have greatly increased. VOD service providers want to be able to provide users with high-quality keyframes of high quality videos within a few minutes after the broadcast ends. However, existing keyframe extraction tends to select keyframes whose quality as a keyframe is insufficiently considered, and it takes a long computation time because it does not consider an HD class image. In this paper, we propose a keyframe selection method that flexibly applies multiple keyframe quality metrics and improves the computation time. The main procedure is as follows. After shot boundary detection is performed, the first frames are extracted as initial keyframes. The user sets evaluation metrics and priorities by considering the genre and attributes of the video. According to the evaluation metrics and the priority, the low-quality keyframe is selected as a replacement target. The replacement target keyframe is replaced with a high-quality frame in the shot. The proposed method was subjectively evaluated by 23 votes. Approximately 45% of the replaced keyframes were improved and about 18% of the replaced keyframes were adversely affected. Also, it took about 10 minutes to complete the summary of one hour video, which resulted in a reduction of more than 44.5% of the execution time.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제16권5호
• /
• pp.1755-1777
• /
• 2022

The development of wireless communication technology has led to the underutilization of radio spectra. To address this limitation, an intelligent cognitive radio network was developed. Specific emitter identification (SEI) is a key technology in this network. However, in realistic non-cooperative scenarios, the system may detect signal classes beyond those in the training database, and only a few labeled signal samples are available for network training, both of which deteriorate identification performance. To overcome these challenges, a meta-learning-based open-set identification system is proposed for SEI. First, the received signals were pre-processed using bi-spectral analysis and a Radon transform to obtain signal representation vectors, which were then fed into an open-set SEI network. This network consisted of a deep feature extractor and an intrinsic feature memorizer that can detect signals of unknown classes and classify signals of different known classes. The training loss functions and the procedures of the open-set SEI network were then designed for parameter optimization. Considering the few-shot problems of open-set SEI, meta-training loss functions and meta-training procedures that require only a few labeled signal samples were further developed for open-set SEI network training. The experimental results demonstrate that this approach outperforms other state-of-the-art SEI methods in open-set scenarios. In addition, excellent open-set SEI performance was achieved using at least 50 training signal samples, and effective operation in low signal-to-noise ratio (SNR) environments was demonstrated.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제16권4호
• /
• pp.1166-1186
• /
• 2022

Artistic font design has become an integral part of visual media. However, without prior knowledge of the font domain, it is difficult to create distinct font styles. When the number of characters is limited, this task becomes easier (e.g., only Latin characters). However, designing CJK (Chinese, Japanese, and Korean) characters presents a challenge due to the large number of character sets and complexity of the glyph components in these languages. Numerous studies have been conducted on automating the font design process using generative adversarial networks (GANs). Existing methods rely heavily on reference fonts and perform font style conversions between different fonts. Additionally, rather than capturing style information for a target font via multiple style images, most methods do so via a single font image. In this paper, we propose a network architecture for generating multilingual font sets that makes use of geometric structures as content. Additionally, to acquire sufficient style information, we employ multiple style images belonging to a single font style simultaneously to extract global font style-specific information. By utilizing the geometric structural information of content and a few stylized images, our model can generate an entire font set while maintaining the style. Extensive experiments were conducted to demonstrate the proposed model's superiority over several baseline methods. Additionally, we conducted ablation studies to validate our proposed network architecture.

• 대한기계학회논문집B
• /
• 제21권12호
• /
• pp.1679-1689
• /
• 1997

In this paper, a precise description is given to the basic theory as well as the detailed algorithms for the numerical treatment of the method of POD (proper orthogonal decomposition). This method is then applied to analysing the numerical solutions of one-dimensional shallow-water equations to show how the method is affected by various parameters such as the sampling time, sampling numbers, and the spatial resolution for the autocorrelation function. A few curious features associated with this flow model found through the analysis are further explained and discussed.

• 한국정보과학회 언어공학연구회:학술대회논문집(한글 및 한국어 정보처리)
• /
• 한국정보과학회언어공학연구회 2022년도 제34회 한글 및 한국어 정보처리 학술대회
• /
• pp.77-81
• /
• 2022

최근 사전학습 언어모델에 내재된 지식을 최대한으로 활용하고자 태스크에 대한 설명을 입력으로 주는 manual prompt tuning 방법과 자연어 대신 학습가능한 파라미터로 태스크에 대한 이해를 돕는 soft prompt tuning 방법론이 자연어처리 분야에서 활발히 연구가 진행되고 있다. 이에 본 연구에서는 페르소나 대화 생성 태스크에서 encoder-decoder 구조 기반의 사전학습 언어모델 BART를 활용하여 manual prompt tuning 및 soft prompt tuning 방법을 고안하고, 파인튜닝과의 성능을 비교한다. 전체 학습 데이터에 대한 실험 뿐 아니라, few-shot 세팅에서의 성능을 확인한다.