• 한국멀티미디어학회:학술대회논문집
• /
• 한국멀티미디어학회 2012년도 춘계학술발표대회논문집
• /
• pp.266-266
• /
• 2012

• 정보처리학회논문지C
• /
• 제9C권5호
• /
• pp.757-764
• /
• 2002

본 논문에서는 분산 멀티미디어 응용을 위한 객체 지향형 프레임워크 방식으로 설계된 네트워크 멀티미디어 관리 시스템 개발에 관한 설계 이슈 및 프로토타입 개발 경험을 기술하고, 특히 멀티미디어 관리 시스템의 일부로서 개발한 비디오 서버 상에 적용 가능한 통합형 QoS-자원 협상 프로토콜의 주요 특징에 관하여 설명한다. 구현된 멀티미디어 프레임 워크 상에서 효율적인 멀티미디어 스트리밍을 지원하기 위하여 새로 제안된 QoS 협상 정책을 실제 서버에 적용하는데 있어서 고려해야 할 중요 이슈에 대해서도 분석하였다.

• 한국정보처리학회논문지
• /
• 제7권2S호
• /
• pp.678-691
• /
• 2000

A variety of multimedia applications require multimedia database management systems to manage multimedia data, such as text, image, and video, as well as t support content-based image or video retrieval. In this paper we design and implement a content-based image retrieval system, BADA-IV/I$^2$R(Image Information Retrieval), which is developed based on BADA-IV multimedia database management system. In this system image databases can be efficiently constructed and retrieved with the visual features, such as color, shape, and texture, of image. we extend SQL statements to define image query based on both annotations and visual features of image together. A high-dimensional index structure, called CIR-tree, is also employed in the system to provide an efficient access method to image databases. We show that BADA-IV/I$^2$R provides a flexible way to define query for image retrieval and retrieves image data fast and effectively: the effectiveness and performance of image retrieval are shown by BEP(Bull's Eye Performance) that is used to measure the retrieval effectiveness in MPEG-7 and comparing the performance of CIR-tree with those of X-tree and TV-tree, respectively.

• ETRI Journal
• /
• 제44권1호
• /
• pp.38-50
• /
• 2022

This paper presents a study on the use of encoder-derived features in decoder-side depth estimation. The scheme of multiview video encoding does not require the transmission of depth maps (which carry the geometry of a three-dimensional scene) as only a set of input views and their parameters are compressed and packed into the bitstream, with a set of features that could make it easier to estimate geometry in the decoder. The paper proposes novel recursive block splitting for the feature extraction process and evaluates different scenarios of feature-driven decoder-side depth estimation, performed by assessing their influence on the bitrate of metadata, quality of the reconstructed video, and time of depth estimation. As efficient encoding of multiview sequences became one of the main scopes of the video encoding community, the experimental results are based on the "geometry absent" profile from the incoming MPEG Immersive video standard. The results show that the quality of synthesized views using the proposed recursive block splitting outperforms that of the state-of-the-art approach.

• 한국멀티미디어학회논문지
• /
• 제7권10호
• /
• pp.1452-1463
• /
• 2004

양방향 방송 서비스로의 전환을 맞아 다양한 사용자 요구 및 기호에 적합한 컨텐츠를 제공하고, 증가하는 방송 컨텐츠를 효율적으로 관리, 이용하기 위해 비디오의 자동 에 대한 요구가 증가하고 있다. 본 논문에서는 내용 구성이 잘 갖추어진 어학 교육 비디오의 자동 에 대한 방법을 제안한다. 내용 기반을 자동으로 생성하기 위해 먼저 디지털 비디오로부터 샷 경계를 검출한 후, 각 샷을 대표하는 키프레임으로부터 비주얼 특징들을 추출한다. 그리고 추출된 다중 비주얼 특징을 이용해 어학 교육 비디오의 세분화된 내용 정보를 결정한다. 마지막으로, 결정된 내용 정보를 기술하는 요약문을 MPEG-7 MDS(Multimedia Description cheme)에 정의된 계층적 (Hierarchical Summary) 구조에 맞추어 XML 문서로 생성한다. 외국어 회화 비디오에 대해 실험하여 제안한 자동 방법의 효율성을 검증하였으며, 제안한 방법이 교육 방송용 컨텐츠의 다양한 서비스 제공 및 관리를 위한 비디오 요약 시스템에 효율적으로 적용 가능함을 확인하였다.

• 한국멀티미디어학회논문지
• /
• 제18권5호
• /
• pp.610-619
• /
• 2015

*In this paper, we propose a real-time face tracking system using an adaptive face detector and a tracking algorithm. An image is divided into the regions of background and face candidate by a real-time updated skin color identifying system in order to accurately detect facial features. The facial characteristics are extracted using the five types of simple Haar-like features. The extracted features are reinterpreted by Principal Component Analysis (PCA), and the interpreted principal components are processed by Support Vector Machine (SVM) that classifies into facial and non-facial areas. The movement of the face is traced by Kalman filter and Mean shift, which use the static information of the detected faces and the differences between previous and current frames. The proposed system identifies the initial skin color and updates it through a real-time color detecting system. A similar background color can be removed by updating the skin color. Also, the performance increases up to 20% when the background color is reduced in comparison to extracting features from the entire region. The increased detection rate and speed are acquired by the usage of Kalman filter and Mean shift.

• 한국멀티미디어학회논문지
• /
• 제23권8호
• /
• pp.1019-1029
• /
• 2020

In this paper, we propose a novel knowledge distillation algorithm to create an compressed deep ensemble network coupled with the combined use of local and global features of face images. In order to transfer the capability of high-level recognition performances of the ensemble deep networks to a single deep network, the probability for class prediction, which is the softmax output of the ensemble network, is used as soft target for training a single deep network. By applying the knowledge distillation algorithm, the local feature informations obtained by training the deep ensemble network using facial subregions of the face image as input are transmitted to a single deep network to create a so-called compressed ensemble DCNN. The experimental results demonstrate that our proposed compressed ensemble deep network can maintain the recognition performance of the complex ensemble deep networks and is superior to the recognition performance of a single deep network. In addition, our proposed method can significantly reduce the storage(memory) space and execution time, compared to the conventional ensemble deep networks developed for face recognition.

• 한국멀티미디어학회논문지
• /
• 제22권12호
• /
• pp.1457-1465
• /
• 2019

3D convolution is to stack multiple consecutive frames to form a cube, and then apply the 3D convolution kernel in the cube. In this structure, each feature map of the convolutional layer is connected to multiple adjacent sequential frames in the previous layer, thus capturing the motion information. However, due to the changes of pedestrian posture, motion and position, the convolution at the same place is inappropriate, and when the 3D convolution kernel is convoluted in the time domain, only time domain features of three consecutive frames can be extracted, which is not a good enough to get action information. This paper proposes an action recognition method based on feature fusion of 3D convolutional neural network. Based on the VGG16 network model, sending a pre-acquired optical flow image for learning, then get the time domain features, and then the feature of the time domain is extracted from the features extracted by the 3D convolutional neural network. Finally, the behavior classification is done by the SVM classifier.

• Journal of Multimedia Information System
• /
• 제7권2호
• /
• pp.97-106
• /
• 2020

The subjective experience would degrade the current artificial artistic aesthetic analysis. Since Qiong kiln ceramics have a long history and occupy a very important position in ceramic arts, we employed computer-aided technologies to quickly automatically accurately and quantitatively process a large number of Qiong kiln ceramic images and generate the detailed statistical data. Because the color features are simple and significant visual characteristics, the color features of Qiong kiln ceramics are analyzed for the quantitative aesthetics. The Qiong kiln ceramic images are segmented with GrabCut algorithm. Three moments (1st-order, 2nd-order, and 3rd-order) are calculated in two typical color spaces, namely RGB and HSV. The discrimination powers of the color features are analyzed according to various dynasties (Tang Dynasty, Five Dynasties, Song Dynasty) and various utensils (Pot, kettle, bowl), which are helpful to the selection of the discriminant color features among various dynasties and utensils. This paper is helpful to promoting the quantitative aesthetic research of Qiong kiln ceramics and is also conducive to the research on the aesthetics of other ceramics.

• 한국멀티미디어학회논문지
• /
• 제5권5호
• /
• pp.505-514
• /
• 2002

본 논문에서는 내용기반 영상 분류를 위한 방법론으로써 신경망을 이용한 방법을 제안한다. 분류 대상 영상은 인터넷상의 다양한 영상들 중에서 전경과 배경의 구분이 있는 객체 영상이다. 전처리 과정에서 영역 분할을 이용하여 영상 내에서 배경을 제거하고 객체 영역을 추출한다. 분류를 위한 특징은 웨이블릿 변환 후 푸출된 형태 특징과 질감특징을 이용한다 추출된 특징 값들을 이용하여 영상들에 대한 학습패턴을 생성하고 신경망 분류기를 구성 한다. 신경망의 학습 알고리즘은 역전파 알고리즘을 사용한다. 가장 효과적인 질감특징을 선 택 하기 위한 실험에서는 대각 모멘트가 가장 높은 분류률을 보여 주었다. 배경을 제거 하고 대각 모멘트를 특징으로 사용하여 실험하였을 때, 30종류에서 각 10개씩 총 300개의 학습 데이터와300개의 테스트 데이터에 대하여 각각 72.3%와 67%의 정분류률을 보였다.