서지주요정보
VA 파일 성능의 분석과 향상 = Analysis and improvement of the VA-file performance
서명 / 저자 VA 파일 성능의 분석과 향상 = Analysis and improvement of the VA-file performance / 최승락.
저자명 최승락 ; Choi, Seung-Lak
발행사항 [대전 : 한국과학기술원, 1999].
Online Access 원문보기 원문인쇄

소장정보

등록번호

8009846

소장위치/청구기호

학술문화관(문화관) 보존서고

MCS 99042

SMS전송

도서상태

이용가능

대출가능

반납예정일

등록번호

9006031

소장위치/청구기호

서울 학위논문 서가

MCS 99042 c. 2

SMS전송

도서상태

이용가능

대출가능

반납예정일

초록정보

Recently, similarity search has been playing an important role in multimedia databases. However, a conventional approach for similarity search, partitioning index structures which partition vector spaces recursively, perform poorly as dimensionality increases. Moreover, even a linear scan outperforms the partitioning index structures when the dimensionality exceeds around ten. This phenomenon is often called the `dimensionality curse.' The VA-file, that is proposed as an alternative to the partitioning index structures, is a flat file containing approximations for object vectors, and is used as a filter. The VA-file retains good performance in high dimensionality. In this paper, we present a generalization of similarity search algorithms of the partitioning index structures and analyze the generalized algorithm. From the analysis, we find out a fact that the dimensionality curse arises from long diameters of minimum bounding regions (MBRs) in high dimensional vector spaces. We then provide a cost model for the similarity search in the VA-file. The performance of the VA-file varies as a function of the number of bits b, used for an approximation. Therefore, it is very important to determine b in order to optimize the performance of the VA-file. Finally, we analyze the performance behavior of the VA-file, and suggest the guideline to determine b that produces the best performance.

서지기타정보

서지기타정보
청구기호 {MCS 99042
형태사항 33 p. : 삽도 ; 26 cm
언어 한국어
일반주기 저자명의 영문표기 : Seung-Lak Choi
지도교수의 한글표기 : 이윤준
지도교수의 영문표기 : Yoon-Joon Lee
학위논문 학위논문(석사) - 한국과학기술원 : 전산학과,
서지주기 참고문헌 : p. 31-33
주제 멀티미디어 데이터베이스
유사성 검색
k-최근접 검색
VA 파일
비용 모델
Multimedia database
Similarity search
k-nearest neighbor search
VA-file
Cost model
QR CODE qr code