公开(公告)号：KR1020010057841A

公开(公告)日：2001-07-05

申请号：KR1019990061254

申请日：1999-12-23

Applicant: 한국전자통신연구원

Inventor： 이재연 ,  정세윤 ,  김규헌 ,  전병태 ,  배영래

IPC: G06F17/30

Abstract: PURPOSE: A method for searching an image using a representative value of a high-dimensional image feature volume is provided to efficiently increase the speed and the efficiency of searching by extracting a representative value from a feature volume vector, and by restricting a range of the feature volume of a database to be searched. CONSTITUTION: A characteristic extracting unit extracts a characteristic volume of a query language from a query image(300). A search range restriction unit receives the characteristic volume, and extracts one or more representative values(302). An image database selects a plurality of candidate sets(304). The search range restriction unit forms a plurality of candidate sets selected in the form of bit array(306). The search range restriction unit performs a bitwise AND calculation of a plurality of candidate sets(308). A similarity estimation unit estimates a similarity of characteristic volume of the candidate sets, and searches an image corresponded to the query image(310).

Abstract translation: 目的：提供使用高维图像特征量的代表值来搜索图像的方法，以通过从特征体积向量提取代表值来有效地提高搜索的速度和效率，并且通过限制 要搜索的数据库的特征量。 构成：特征提取单元从查询图像提取查询语言的特征量（300）。 搜索范围限制单元接收特征体积，并提取一个或多个代表值（302）。 图像数据库选择多个候选集（304）。 搜索范围限制单元形成以位阵列形式选择的多个候选集（306）。 搜索范围限制单元执行多个候选集合的按位AND运算（308）。 相似度估计单元估计候选集的特征体积的相似度，并且搜索与查询图像相对应的图像（310）。

公开(公告)号：KR1020010057807A

公开(公告)日：2001-07-05

申请号：KR1019990061217

申请日：1999-12-23

Applicant: 한국전자통신연구원

Inventor： 김규헌 ,  정세윤 ,  전병태 ,  이재연 ,  배영래

IPC: G06T1/00

Abstract: PURPOSE: A segmentation picture postprocessing method using Quasi third Markov random field is provided to realize an accurate segmentation and classification by correcting region level of central pixel and using Quasi third neighboring system. CONSTITUTION: A neighboring system composed of small window is defined to get more region level information of surround pixel using Quasi third Markov random field(MRF). The relation between central pixel and surround pixel of the neighboring system defined by the Quasi third Markov random field is expressed by each conditional probability function under condition of each different region level. The region level supplying the highest probability among these is corrected by central pixel of an appropriate window. A whole picture which is made a segmentation including error by overlapping each window each other is corrected.

Abstract translation: 目的：提供使用准第三马尔科夫随机场的分割图像后处理方法，通过校正中心像素的区域水平和使用准第三相邻系统来实现精确的分割和分类。 构成：使用准第三马尔科夫随机场（MRF）定义由小窗口组成的相邻系统以获得更多的环绕像素的区域级信息。 由准第三马尔可夫随机场定义的相邻系统的中心像素和环绕像素之间的关系由每个不同区域级别的条件下的每个条件概率函数表示。 在这些中提供最高概率的区域级由相应窗口的中心像素校正。 校正通过使每个窗口彼此重叠而构成包括错误的分割的整个图像。

公开(公告)号：KR1020010004402A

公开(公告)日：2001-01-15

申请号：KR1019990025033

申请日：1999-06-28

Applicant: 한국전자통신연구원

Inventor： 김규헌 ,  정세윤 ,  전병태 ,  이재연 ,  배영래

IPC: G06F17/30

Abstract: PURPOSE: A texture-based searching method with co-occurrence matrix and normalized Euclidean distance is provided to search a moving image effectively based on co-occurrence matrix and normalized Euclidean distance. CONSTITUTION: A texture feature vector is obtained through co-occurrence matrix of a selected distance of a sample image. The input image is analyzed based on a window. A texture feature is obtained by the co-occurrence matrix of the analyzed window-based image. A normalized Euclidean distance between the texture feature vectors of the selected distance and the window is obtained. The normalized Euclidean distance is compared with a specific value. Finally, the current texture is recognized as a different one if the normalized Euclidean distance is larger or the same as the specific value. Otherwise, the texture is the same.

Abstract translation: 目的：提出了一种基于纹理的搜索方法，具有同现矩阵和归一化欧几里德距离，用于基于同现矩阵和归一化欧几里德距离有效搜索运动图像。 构成：通过样本图像的选定距离的同现矩阵获得纹理特征向量。 基于窗口分析输入图像。 通过分析的基于窗口的图像的共生矩阵获得纹理特征。 获得所选距离的纹理特征向量与窗口之间的归一化欧氏距离。 将归一化的欧氏距离与特定值进行比较。 最后，如果归一化的欧氏距离较大或与特定值相同，则当前纹理被识别为不同的纹理。 否则，纹理是一样的。

公开(公告)号：KR1019970017001A

公开(公告)日：1997-04-28

申请号：KR1019950030555

申请日：1995-09-15

Applicant: 한국전자통신연구원

Inventor： 최경호 ,  양영규 ,  배영래 ,  이종훈 ,  조성익 ,  강태호 ,  김경옥 ,  정인숙 ,  김진서 ,  정동석

IPC: G06T1/00

Abstract: 본 발명은 자동시각검사 시스템(Automated Visual Inspection System)의 구성에 필요한 시각검사 알고리즘을 생성시키도록 하는 자동시각검사용 알고리즘 개발방법 및 장치에 관한 것으로 시각검사 알고리즘의 개발에 다수의 숙련된 프로그램 개발자들과 시간이 필요하게 되고 검사할 대상물이 바뀔때마다 새로운 시각검사 알고리즘을 개발하여야 하는 문제점을 없앤 것이다.
이러한 본 발명은 조작자가 CCD 카메라의 앞에 검사 대상물을 위치 시키거나 촬영된 대상물의 비데오 테이프르 비데오 플레이어에 입력시키고 퍼스널 컴퓨터를 조작하면, 이 퍼스널 컴퓨터의 버스라인보드를 통해 전송되며, 전송된 영상을 영상처리 알고리즘과 그래픽 알고리즘을 이용하여 대상물의 특징추출에 적당한 영상으로 변환시키는 전처리 단계를 수행하고, 처리된 영상으로부터 특징 데이타를 추출하는 영역결정 단계과 대상물의 결합여부 판단을 위한 결정규친 단계를 거쳐서 시각검사 알고리즘을 생성시키게 된다.