• Journal of KIISE:Computer Systems and Theory
• /
• v.31 no.9
• /
• pp.527-535
• /
• 2004

Handling mixed workload in digital set-top box or streaming server becomes an important issue as integrated file system gets momentum as the choice for the next generation file system. The next generation file system is required to handle real-time audio/video playback while being able to handle text requests such as web page, image file, etc. Legacy file system provides only best effort I/O service and thus cannot properly support the QoS of soft real-time I/O. In this paper, we would like to present our experience in developing the file system which fan guarantee the QoS of multimedia stream. We classify all application I/O requests into two category: periodic I/O and sporadic I/O. The QoS requirement of multimedia stream could be guaranteed by giving a higher priority to periodic requests than sporadic requests. The proto-type file system(Qosfs) is developed on Linux Operating System.

• Journal of KIISE:Computing Practices and Letters
• /
• v.6 no.1
• /
• pp.82-92
• /
• 2000

This paper introduces an implementation of the diagnosis software for CC-NUMA systems. The CC-NUMA architecture is composed of two or more SMP nodes installed with the specialized hardware to provide cache-coherent operation and the high-speed interconnection network to connect each node, it enables both the high performance and the high scalability. While the CC-NUMA system provides the single system image in the operating system aspect, it should be considered the multiple systems by the diagnostic software. Thus it is difficult to diagnose and manage CC-NUMA system using commercial administration software due to characteristics of the complicated architecture. The remote diagnosis and management are also required with a view to reduce Total Cost of Ownership. In this paper, we design diagnostic software to manage CC-NUMA server system, and propose its mechanism in client-server manner to support remote administration. Additionally, we use the Java-based user interface to enlarge an administrator's accessibility.

• Journal of KIISE
• /
• v.44 no.1
• /
• pp.44-50
• /
• 2017

Pedestrian detection, an important component of autonomous driving and driving assistant system, has been extensively studied for many years. In particular, image based pedestrian detection methods such as Hierarchical classifier or HOG and, deep models such as ConvNet are well studied. The evaluation score has increased by the various methods. However, pedestrian detection requires high sensitivity to errors, since small error can lead to life or death problems. Consequently, further reduction in pedestrian detection error rate of autonomous systems is required. We proposed a new method to detect pedestrians and reduce the error rate by using the Faster R-CNN with new developed pedestrian training data sets. Finally, we compared the proposed method with the previous models, in order to show the improvement of our method.

• Journal of KIISE:Software and Applications
• /
• v.27 no.3
• /
• pp.201-209
• /
• 2000

Recent works regarding video shot change detection algorithms show that abrupt shot changes are detected fairly well. However, gradual shot changes including wipes and dissolves are often missed or falsely detected. A robust shot change detection system, therefore, must include a shot verification step to further enhance the overall system performance. In this paper, we introduce the concept of the visual rhythm which is a single image, a subsampled version of a full video. On the visual rhythm, the different video edit effects such as cuts, wipes and dissolves manifest themselves as different patterns. Using this characteristic, it becomes possible, without sequentially playing the entire video, to find false positive shots as well as undetected shots. Thus, inclusion of the visual rhythm in the shot boundary verification process will aid the operator to exclude falsely detected shots as well as to find undetected shots fast and efficiently. For this purpose we have developed a new tool, a shot verifier incorporating the visual rhythm. The usefulness of the visual rhythm during the shot verification process will be presented.

• Journal of KIISE
• /
• v.43 no.11
• /
• pp.1281-1297
• /
• 2016

In recent years, personal videos have seen a tremendous growth due to the substantial increase in the use of smart devices and networking services in which users create and share video content easily without many restrictions. However, taking both into account would significantly improve event detection performance because videos generally have multiple modalities and the frame data in video varies at different time points. This paper proposes an event detection method. In this method, high-level features are first extracted from multiple modalities in the videos, and the features are rearranged according to time sequence. Then the association of the modalities is learned by means of DNN to produce a personal video event detector. In our proposed method, audio and image data are first synchronized and then extracted. Then, the result is input into GoogLeNet as well as Multi-Layer Perceptron (MLP) to extract high-level features. The results are then re-arranged in time sequence, and every video is processed to extract one feature each for training by means of DNN.

• Journal of KIISE:Computer Systems and Theory
• /
• v.27 no.3
• /
• pp.235-244
• /
• 2000

This paper addresses a new method for constructing surface representation of 3D structures from a sequence of tomographic cross-sectional images, Firstly, we propose cell-boundary representation by transforming the cuberille space into cell space. A cell-boundary representation consists of a set of boundary cells with their 1-voxel configurations, and can compactly describe binary volumetric data. Secondly, to produce external surface from the cell-boundary representation, we define 19 modeling primitives (MP) including volumetric, planar and linear groups. Surface polygons are created from those modeling primitives using a simple table look-up operation. Comparing with previous method such as Marching Cube or PVP algorithm, our method is robust and does not make any crack in resulting surface model. Hardware implementation is expected to be easy because our algorithm is simple(scan-line), efficient and guarantees data locality in computation time.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.6
• /
• pp.654-662
• /
• 2010

In this paper, we suggest an advanced algorithm, to recognize pedestrian/non-pedestrian using second-stage cascade method, which applies Adaboost algorithm to make a strong classification from weak classifications. First, we extract two feature vectors: (i) Histogram of Oriented Gradient (HOG) which includes gradient information and differential magnitude; (ii) Curvature-HOG which is based on four different curvature features per pixel. And then, a strong classification needs to be obtained from weak classifications for composite recognition method using both HOG and curvature-HOG. In the proposed method, we use one feature vector and one strong classification for the first stage of recognition. For the recognition-failed image, the other feature and strong classification will be used for the second stage of recognition. Based on our experiment, the proposed algorithm shows higher recognition rate compared to the traditional method.

• Journal of KIISE:Information Networking
• /
• v.36 no.3
• /
• pp.198-203
• /
• 2009

With the growing demand for multimedia services and advances in display technology, new applications for 3$\sim$D scene communication have emerged. While multi-view video of these emerging applications may provide users with more realistic scene experience, drastic increase in the bandwidth is a major problem to solve. In this paper, we propose a fast prediction mode decision algorithm which can significantly reduce complexity and time consumption of the encoding process. This is based on the object segmentation, which can effectively identify the fast moving foreground object. As the foreground object with fast motion is more likely to be encoded in the view directional prediction mode, we can properly limit the motion compensated coding for a case in point. As a result, time savings of the proposed algorithm was up to average 45% without much loss in the quality of the image sequence.

• Proceedings of the Korean Information Science Society Conference
• /
• 2010.06c
• /
• pp.31-36
• /
• 2010

최근 대두되고 있는 웹2.0의 특징은 일반 사용자들이 능동적으로 정보를 생산해내고 공유하는데 있다. 웹 2.0의 참여형 아키텍쳐를 구성하는 핵심요소로 인식되고 있는 폭소노미(Folksonomy)는 과거 택소노미(Taxonomy)와 같이 전문가에 의하여 구축되는 분류 체계가 아닌 사용자들이 협동적으로 태그(Tag)들을 만들고 관리하는 소셜 태깅(Social Tagging)에 의한 분류 시스템이다. 최근 이러한 폭소노미를 활용하여 이미지를 공유하고 검색하고자 하는 다양한 시도들이 진행되고 있다. 그러나 Flickr와 같은 태그 기반 이미지 공유 시스템에서는 태그의 문법적, 의미적 모호성과 이미지에 대한 태그들의 중요성 또는 상관관계를 고려하지 않아 태그 기반 검색 시 정확성 및 신뢰성을 보장할 수 없다. 이러한 문제를 해결하기 위해 폭소노미에 기반한 이미지 공유 데이터베이스에서 적합한 태그들을 태그 전달(Tag Propagation)하거나 확률 및 출현빈도에 기반하여 태그 랭킹을 수행하기 위한 연구들이 활발히 진행되고 있지만 여전히 만족할만한 성능을 보이지 못하고 있다. 본 논문에서는 이미지 공유 데이터베이스에서 유사한 이미지들로부터 이미지에 보다 적합한 태그들을 부여하기 위해서, WordNet을 활용하여 태그들 간의 의미관계에 기반한 효율적인 태그 랭킹 기법을 제안한다. 또한, 신뢰성 있는 태그 기반 검색을 위하여 제안한 태그 랭킹 기법이 현재 이미지 공유 시스템의 랭킹 결과보다 정확성을 높일 수 있음을 실험 예제를 통하여 확인하였다.

• Journal of KIISE:Databases
• /
• v.32 no.4
• /
• pp.362-371
• /
• 2005

Similarity indexing and searching are well known to be difficult in high-dimensional applications such as multimedia databases. Especially, they become more difficult when multiple features have to be indexed together. In this paper, we propose a novel indexing method called the GB-index that is designed to efficiently handle complex similarity queries as well as relevance feedback in high-dimensional image databases. In order to provide the flexibility in controlling multiple features and query objects, the GB-index treats each dimension independently The efficiency of the GB-index is realized by specialized bitmap indexing that represents all objects in a database as a set of bitmaps. Main contributions of the GB-index are three-fold: (1) It provides a novel way to index high-dimensional data; (2) It efficiently handles complex similarity queries; and (3) Disjunctive queries driven by relevance feedback are efficiently treated. Empirical results demonstrate that the GB-index achieves great speedups over the sequential scan and the VA-file.