• Proceedings of the Korea Multimedia Society Conference
• /
• 2000.11a
• /
• pp.256-261
• /
• 2000

디지털 방송, 웹의 발전으로 내용 기반 검색, 비디오 인덱싱, 비디오 검색 등의 시스템들이 많이 연구, 개발되고 있으며, 이러한 시스템에서는 사람을 주제로 검색하는 요구가 많이 발생한다. 대부분의 얼굴 영역 추출 및 인식 시스템들은 질감, 모양, 움직임, 칼라 등의 특징들을 이용하는데, 이들 중 칼라 특징은 기존 시스템의 첫 번째 처리 단계에서 많이 사용된다. 하지만, 복잡한 배경, 조명, 화장(make up), 잡영들 때문에 미리 정의된 단일 칼라 임계값을 이용하여 얼굴 영역과 비 얼굴 영역으로 구분하면 정확한 추출 결과를 얻기 힘들다는 문제가 있다. 본 논문에서는, 점진적으로 피부색 영역을 확장시키면서 얼굴 영역을 추출하는 방법을 제안한다. 이때 확장 단계에서 얼굴 영역을 판단하기 위해, 일굴 각 기관들의 위치적 정보를 사용하였다. 얼굴 기관은 눈과 입을 사용했는데, 여러 가지 요인으로 인해 이들을 정확하게 추출하기가 어렵기 때문에, 각 단계에서 얼굴 후보 영역 내부의 수평 방향성을 가지는 경계를 눈과 입의 영역으로 간주했다. 실험을 통해, 제안한 방법이 하이라이트(highlight)에 의해 얼굴 영역의 일부가 왜곡된 경우와 얼굴 영역이 피부색과 유사한 배경에 인접해 있는 경우에 대해서도 강인하게 얼굴 영역을 추출할 수 있음을 확인하였다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.01a
• /
• pp.93-95
• /
• 2023

본 논문에서는 장편의 드라마나 영화에서 스토리 기반의 축약된 요약본을 자동으로 제작하기 위해 미디엄 숏(medium shot) 크기의 숏(shot)들을 추출하기 위한 방법을 고려한다. 미디엄 숏 정도의 크기는 보통 인물에 중심을 둔 숏들로 인물들 간의 관계에서 특히 대사나 표정으로 내용을 전달하기 위한 목적으로 적극 권장된다. 비디오 검색을 위한 인덱싱에서 신(scene) 전환 검출 및 숏 경계 검출, 그리고 이미지에서 심도와 초점기반의 화질 및 피사체 추출 등을 위해 전통적인 신호/영상처리 기법의 활용에서부터 최근의 기계학습 접목 등 다양한 연구들이 진행되고 있다. 영상문법에 근거하여 편집된 영상물에서 미디엄 숏 정도 크기의 숏들을 추출하여 배열한다면 어느 정도 원본 내용을 충실히 전달할 수 있는 축약된 요약본을 제작할 수 있다는 가정하에 해당 샷들을 블러(blur) 기반으로 검출하기 위해 이와 관련된 키워드들을 기반으로 기존 연구들을 살펴보고 적용 방법을 모색한다.

• Journal of KIISE:Software and Applications
• /
• v.30 no.7_8
• /
• pp.727-743
• /
• 2003

This paper proposes an automatic highlight building algorithm for soccer video by using the structural characteristics of broadcasted sports video that an interesting (or important) event (such as goal or foul) in sports video has a continuous replay shot surrounded by gradual shot change effect like wipe. This shot editing rule is used in this paper to analyze the structure of broadcated soccer video and extracts shot involving the important events to build a highlight. It first uses the spatial-temporal image of video to detect wipe transition effects and zoom out/in shot changes. They are used to detect the replay shot. However, using spatial-temporal image alone to detect the wipe transition effect requires too much computational resources and need to change algorithm if the wipe pattern is changed. For solving these problems, a two-pass detection algorithm and a pixel sub-sampling technique are proposed in this paper. Furthermore, to detect the zoom out/in shot change and replay shots more precisely, the green-area-ratio and the motion energy are also computed in the proposed scheme. Finally, highlight shots composed of event and player shot are extracted by using these pre-detected replay shot and zoom out/in shot change point. Proposed algorithm will be useful for web services or broadcasting services requiring abstracted soccer video.

• Journal of Broadcast Engineering
• /
• v.15 no.6
• /
• pp.853-867
• /
• 2010

Recently, thumbnail techniques are required to index a high-performance video at digital convergence-based multimedia service like IPTV and DMB. Therefore a thumbnail extraction method in H.264/AVC bitstreams has been proposed. However, thumbnail quality deterioration problem at converting the general equation of spatial domain to frequency domain which is generated by not considering about H.264/AVC transform and quantization processing and rounding-off operation in intra prediction. In this paper, we propose a new thumbnail extraction method in H.264/AVC bitstreams. The proposed scheme is based on H.264/AVC core-transform for a thumbnail extraction in frequency domain, and probability theory, intra rounding-off error compensation. Through the implementation and performance evaluation, the subjective quality difference between the output of our scheme and the output of reference decoder is negligible and better than the conventional method, and moreover PSNR gain by up to 8.66 dB.

• The Journal of the Korea Contents Association
• /
• v.11 no.10
• /
• pp.59-68
• /
• 2011

As the role of soccer game analyst who analyzes soccer games and creates soccer wining strategies is emphasized, it is required to have high-level analysis beyond the procedural ones such as main event detection in the context of IT based broadcasting soccer game research community. In this paper, we propose a novel approach to generate the high-level in-depth analysis results via real-time text based soccer Webcast and text mining. Proposed method creates a metadata such as attribute, action and event, build index, and then generate available knowledges via text mining techniques such as association rule mining, event growth index, and pathfinder network analysis using Webcast and domain knowledges. We carried out a feasibility experiment on the proposed technique with the Webcast text about Spain team's 2010 World Cup games.

• Journal of Korea Multimedia Society
• /
• v.9 no.9
• /
• pp.1139-1149
• /
• 2006

This paper describes a novel and efficient algorithm, which extracts focused objects from still images with low depth-of-field (DOF). The algorithm unfolds into four modules. In the first module, a HOS map, in which the spatial distribution of the high-frequency components is represented, is obtained from an input low DOF image [1]. The second module finds OOI candidate by using characteristics of the HOS. Since it is possible to contain some holes in the region, the third module detects and fills them. In order to obtain an OOI, the last module gets rid of background pixels in the OOI candidate. The experimental results show that the proposed method is highly useful in various applications, such as image indexing for content-based retrieval from huge amounts of image database, image analysis for digital cameras, and video analysis for virtual reality, immersive video system, photo-realistic video scene generation and video indexing system.

• Journal of KIISE:Software and Applications
• /
• v.33 no.10
• /
• pp.851-861
• /
• 2006

The paper proposes a novel unsupervised video object segmentation algorithm for image sequences with low depth-of-field (DOF), which is a popular photographic technique enabling to represent the intention of photographer by giving a clear focus only on an object-of-interest (OOI). The proposed algorithm largely consists of two modules. The first module automatically extracts OOIs from the first frame by separating sharply focused OOIs from other out-of-focused foreground or background objects. The second module tracks OOIs for the rest of the video sequence, aimed at running the system in real-time, or at least, semi-real-time. The experimental results indicate that the proposed algorithm provides an effective tool, which can be a basis of applications, such as video analysis for virtual reality, immersive video system, photo-realistic video scene generation and video indexing systems.

• The Journal of the Korea Contents Association
• /
• v.11 no.2
• /
• pp.54-60
• /
• 2011

The study is content-based video retrieval system based on the news video domain as researching for the video data processing method for searching the multimedia information. For the implementation of effective system, We retrieval meaning information and special information using the knowing knowledge about formation and structure of the video data. These are possible to retrieval searching by articles fast and accurately by indexing contents the users want to search. The news domain used in experiment of our system is the KBS news on the air nowadays and precision and recall is used to evaluate the experiment and performance.

• Journal of KIISE:Databases
• /
• v.35 no.3
• /
• pp.272-285
• /
• 2008

In this paper, we concentrate on data allocation methods for multiple broadcast channels. When the server broadcasts data, the important issue is to let mobile clients access requested data rapidly. Previous works first sorted data by their access probabilities and allocate the sorted data to the multiple channels by partitioning them into multiple channels. However, they do not reflect the difference of access probabilities among data allocated in the same channel. This paper proposes ZGMD allocation method. ZGMD allocates data item on multiple channels so that the difference of access probability in the same channel is maximized. ZGMD allocates sorted data to each channels and applies Broadcast Disk in each channel. ZGMD requires a proper indexing scheme for the performance improvement. This is because in ZGMD method each channel got allocated both hot and cold data. As a result, the sequential search heuristic does not allow the mobile client to access hot data items quickly. The proposed index scheme is based on using dedicated index channels in order to search the data channel where the requested data is. We show that our method achieve the near-optimal performance in terms of the average access time and significantly outperforms the existing methods.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.20 no.5
• /
• pp.107-112
• /
• 2020

Recently, unstructured data is rapidly being produced based on web-based services. NoSQL systems and key value stores that process unstructured data as key and value pairs are widely used in various applications. In this paper, a study was conducted on a skip list used for in-memory data management in an LSM-tree based key value store. The skip list used in the key value store is an insertion-based skip list that does not allow overwriting and processes all changes only by inserting. This behavior can support Multi-Version Concurrency Control (MVCC), which can simultaneously process multiple read/write requests through snapshot isolation. However, since duplicate keys exist in the skip list, the performance significantly degrades due to unnecessary node visits during a list traverse. In particular, serious overhead occurs when a range query or scan operation that collectively searches a specific range of data occurs. This paper proposes a newly designed Stride SkipList to reduce this overhead. The stride skip list additionally maintains an indexing pointer for the last node of the same key to avoid unnecessary node visits. The proposed scheme is implemented using RocksDB's in-memory component, and the performance evaluation shows that the performance of SCAN operation improves by up to 350 times compared to the existing skip list for various workloads.