-
公开(公告)号:KR1020060070610A
公开(公告)日:2006-06-26
申请号:KR1020040109134
申请日:2004-12-21
Applicant: 한국전자통신연구원
IPC: G06F17/26
CPC classification number: G06F17/30616 , G06F17/218 , G06F17/2223 , G06F17/26 , G06F17/271 , G06F17/2755
Abstract: 본 발명은 질의응답시스템에서의 정답색인 시스템 및 방법에 관한 것으로서, 입력문을 LF(Logical Form) 단위와 문장 단위로 각 어휘들을 정답대상어와 단서어로 나누고, 각 단서어와 정답대상어 쌍을 색인 할 수 있는 방법을 제시하여 정답 제시에서의 속도를 개선함으로써, 질의응답에서 사용자의 질문에 대한 검색 결과인 방대한 문서로부터 사용자가 원하는 정답만을 초점으로 할 수 있으므로 정보검색 및 질의응답 시스템의 성능을 개선할 수 있다.
질의응답, 정답 색인-
公开(公告)号:KR1020050043280A
公开(公告)日:2005-05-11
申请号:KR1020030078115
申请日:2003-11-05
Applicant: 한국전자통신연구원
IPC: G06F17/21
Abstract: 단락 분할 장치 및 방법과 단락분할 방법을 이용한 검색방법이 개시된다. 주제 설정부는 소정의 도메인을 대표하는 학습 데이터에 적어도 하나 이상의 주제를 설정한다. 패턴 추출부는 학습 데이터를 언어분석과정을 통하여 단문으로 분할하고 분할된 각각의 단문으로부터 소정의 어휘쌍으로 이루어진 제1패턴을 추출한다. 패턴 학습부는 추출된 제1패턴을 설정된 주제에 따라 분류하여 주제별 패턴 DB를 생성한다. 단락 분할부는 입력문서를 언어분석과정을 통해 단문으로 분할하고 분할된 각각의 단문으로부터 소정의 어휘쌍으로 이루어진 제2패턴을 추출한 후 제2패턴과 가장 유사한 패턴의 주제를 주제별 패턴 DB로부터 추출하여 단문의 주제를 할당한다. 이로써, e-mail이나 백과사전 문서, 신문 기사와 같은 특정 도메인의 단락별 주제를 파악하는데 용이할 뿐만 아니라 질의 응답이나 정보검색 등의 응용시스템에서 사용자가 원하는 주제만을 선택하여 제공함으로써 검색의 효율을 높일 수 있다.
-
公开(公告)号:KR100446639B1
公开(公告)日:2004-09-04
申请号:KR1020010042482
申请日:2001-07-13
Applicant: 한국전자통신연구원
IPC: G06F17/30
Abstract: PURPOSE: A cell based high dimensional data indexing system and method is provided to index high dimensional data based on a cell for preventing a lowering of a search efficiency in searching for high dimensional data. CONSTITUTION: The method comprises several steps. First, an N dimensional feature vector is extracted from a multimedia object via a feature vector extractor(801). A distance signature is generated via a signature generation module by using a distance between a signature on the feature vector and a cell center(802). One signature is generated by concatenating the feature vector signature and the distance signature(803), and then is stored at a signature database(804). At the same time, the feature vector is stored at a feature vector database(805). A user can perform a search operation on the stored feature vectors by using various queries like a point query, a range query or k-nearest query(806).
Abstract translation: 目的:提供一种基于单元格的高维数据索引系统和方法,用于基于单元索引高维数据,以防止在搜索高维数据时降低搜索效率。 构成:该方法包括几个步骤。 首先,经由特征向量提取器(801)从多媒体对象提取N维特征向量。 经由签名生成模块通过使用特征向量上的签名与小区中心之间的距离来生成距离签名(802)。 通过连接特征向量签名和距离签名生成一个签名(803),然后存储在签名数据库(804)。 同时,特征矢量被存储在特征矢量数据库(805)。 用户可以通过使用诸如点查询,范围查询或k-最近查询(806)的各种查询来对所存储的特征向量执行搜索操作。
-
公开(公告)号:KR1020010036671A
公开(公告)日:2001-05-07
申请号:KR1019990043779
申请日:1999-10-11
Applicant: 한국전자통신연구원
IPC: G06T5/40
Abstract: PURPOSE: A method of generating block-based image histogram is provided to be capable of extracting specific information from a pixel group of a block unit by use of a block of plural pixels by a basic unit of generating histogram. CONSTITUTION: A coefficient 'k' is set to '0', and pixels of an image are grouped to set a block size of a level '0'(S101). Specific information of the image is extracted by a block unit with regard to all blocks of the level '0' to update a corresponding histogram(S102-S105). After extracting the specific information with regard to all blocks of the level '0', the coefficient 'k' indicating the level is increased by '1'(S106). Specific information of a level '1' is generated by merging specific information of the level '0'(S107), updating a corresponding histogram(S108). Operations of extracting the specific information with regard to all blocks of the level '1' and updating a corresponding histogram are repeated(S109,S110). If the operation of generating specific information with regard to all levels is not completed(S111), the procedure goes to the step(S106) of increasing the coefficient. If the operation of generating specific information with regard to all levels is completed(S111), the procedure is ended.
Abstract translation: 目的:提供一种生成基于块的图像直方图的方法,以便能够通过使用多个像素的块通过生成直方图的基本单位从块单元的像素组中提取特定信息。 构成:将系数“k”设置为“0”,并且将图像的像素分组以设置水平“0”的块大小(S101)。 针对电平“0”的所有块,通过块单位提取图像的特定信息,以更新对应的直方图(S102〜S105)。 在提取关于电平“0”的所有块的特定信息之后,指示电平的系数“k”增加“1”(S106)。 通过合并电平“0”的特定信息(S107),更新对应的直方图(S108)来生成电平“1”的具体信息。 重复提取关于电平“1”的所有块的特定信息和更新对应的直方图的操作(S109,S110)。 如果关于所有级别生成特定信息的操作没有完成(S111),则过程进行到增加系数的步骤(S106)。 如果关于所有级别生成特定信息的操作完成(S111),则过程结束。
-
公开(公告)号:KR1020010002385A
公开(公告)日:2001-01-15
申请号:KR1019990022154
申请日:1999-06-15
Applicant: 한국전자통신연구원
IPC: G06T11/00
Abstract: PURPOSE: A method for clustering an image with a duplicate allocation type is provided to perform an image clustering without a drain of memory resource and identify the progress of the clustering in real time. The method performs the clustering to an exemplary image extracted from moving pictures. CONSTITUTION: The method for clustering an image with a duplicate allocation type includes following processes. In a first allocating process, since a newly inputted image(S01) has no the existing cluster, a new cluster is produced(S02) and allocated to the newly inputted image. To the clusters inputted under the second order, a cluster existing the shortest distance to the time axis is searched(S03). Thereafter, a process is to perform whether the distance between the corresponding image and the searched cluster is within a critical value(TH1) or not(S04). If the result is "yes", the corresponding image is allocated to the searched cluster(S05). If the result is "no", a new cluster is produced(S02) and allocated the corresponding image to the cluster.
Abstract translation: 目的:提供一种用于对具有重复分配类型的图像进行聚类的方法,以执行图像聚类,而不会消耗内存资源,并实时识别聚类的进度。 该方法对从运动图像提取的示例性图像执行聚类。 构成:使用重复分配类型对图像进行聚类的方法包括以下过程。 在第一分配处理中,由于新输入的图像(S01)不存在现有的簇,所以产生新的簇(S02)并将其分配给新输入的图像。 对于按二次输入的群集,搜索存在与时间轴的最短距离的群集(S03)。 此后,处理是执行对应图像和搜索到的簇之间的距离是否在临界值(TH1)内(否)(S04)。 如果结果为“是”,则将相应的图像分配给搜索到的集群(S05)。 如果结果为“否”,则生成新的集群(S02),并将相应的映像分配给集群。
-
公开(公告)号:KR1020000060248A
公开(公告)日:2000-10-16
申请号:KR1019990008383
申请日:1999-03-12
Applicant: 한국전자통신연구원
IPC: G06T5/40
Abstract: PURPOSE: A method for forming a multiple histogram using a super pixel and a linear quantization is provided to reduce errors occurring at the edge of colors. CONSTITUTION: The method for forming a multiple histogram using a super pixel and a linear quantization includes following steps. The color, brightness and edge histograms are generated by using color information(Cr, Cb) and brightness information(Y) of compressed image data. The histogram is formed by grouping pixels of the image in one group, that is on a super pixel basis. A macro block is specified on the super pixel basis(S101) in the compressed image and a direct current value of discrete cosine transformation block four brightness(Y) is extracted. Then, whether an edge exists is determined(S102). Finally, a histogram component corresponding to the edge pattern is increased(S104) when an edge exists.
Abstract translation: 目的:提供一种使用超像素和线性量化形成多重直方图的方法,以减少颜色边缘发生的错误。 构成:使用超像素和线性量化形成多个直方图的方法包括以下步骤。 通过使用压缩图像数据的颜色信息(Cr,Cb)和亮度信息(Y)来生成颜色,亮度和边缘直方图。 该直方图是通过将一组中的图像的像素分组,即在超像素的基础上形成的。 在压缩图像中基于超像素(S101)指定宏块,并提取离散余弦变换块四亮度(Y)的直流值。 然后,确定边缘是否存在(S102)。 最后,当边缘存在时,对应于边缘图案的直方图分量增加(S104)。
-
公开(公告)号:KR1020000033366A
公开(公告)日:2000-06-15
申请号:KR1019980050206
申请日:1998-11-23
Applicant: 한국전자통신연구원
IPC: H04N5/225
Abstract: PURPOSE: An apparatus for high speed image acquisition using multi frame memory is provided to enable host computer to read memory of other frame while A/D converter is writing on the frame memory. CONSTITUTION: An apparatus for high speed image acquisition includes a camera, an A/D converter, a multi frame memory block(21), a host computer, and a frame memory selector(22). The camera(10) takes an analog image. The A/D converter converts the analog image signal to digital image data. The multi frame memory block(21) stores the digital image data converted in the A/D converter. The host computer reads the digital image data stored in the multi frame memory block(21) by way of a PCI bus. The frame memory selector selects a frame memory not used by the host computer and acknowledges the frame memory to the A/D converter(21). The frame memory selector(22) selects a frame memory used by the host computer and acknowledges the frame memory to the A/D converter(21).
Abstract translation: 目的:提供一种使用多帧存储器进行高速图像采集的装置,使得主机能够在A / D转换器写入帧存储器时读取其他帧的存储器。 构成:用于高速图像采集的装置包括相机,A / D转换器,多帧存储器块(21),主计算机和帧存储选择器(22)。 照相机(10)拍摄模拟图像。 A / D转换器将模拟图像信号转换为数字图像数据。 多帧存储块(21)存储转换成A / D转换器的数字图像数据。 主机通过PCI总线读取存储在多帧存储块(21)中的数字图像数据。 帧存储器选择器选择主计算机未使用的帧存储器,并向A / D转换器(21)确认帧存储器。 帧存储器选择器(22)选择主计算机使用的帧存储器并向A / D转换器(21)确认帧存储器。
-
-
-
-
-
-