TY - GEN
T1 - Video segment indexing through classification and interactive view-based query
AU - Lee, John Chung Mong
AU - Wei, Xiong
AU - Shen, Ding Gang
AU - Ma, Ruihua
N1 - Publisher Copyright:
© Springer-Verlag Berlin Heidelberg 1996.
PY - 1996
Y1 - 1996
N2 - As video information proliferates, managing video sources becomes increasingly important. Indices must be constructed to allow any future retrieval. We distinguish two categories of indexing: (i) those that are general-purpose and do not make use of domain-specific knowledge, and (ii) those that are application-dependent. In this paper, we present our work in both categories within the VideoBook project. We discuss how to structure video data into shots (physical parts) and clusters (semantic parts). A video partitioning algorithm is described. Its effectiveness and efficiency lies in the use of both statistical and spatial information in the images without, however, having to examine the entire images. To improve the querying efficiency, we propose to investigate in two directions: deriving higher-level indices through classification and providing a method that finds targets of interest through interactive learning. The first technique takes advantage of domain knowledge of underlying applications. The second technique accounts for quantification effect and noise in images and accommodates “learning from negative examples”, resulting into quite good discriminating power. Experimental results are given to demonstrate the effectiveness of our approach.
AB - As video information proliferates, managing video sources becomes increasingly important. Indices must be constructed to allow any future retrieval. We distinguish two categories of indexing: (i) those that are general-purpose and do not make use of domain-specific knowledge, and (ii) those that are application-dependent. In this paper, we present our work in both categories within the VideoBook project. We discuss how to structure video data into shots (physical parts) and clusters (semantic parts). A video partitioning algorithm is described. Its effectiveness and efficiency lies in the use of both statistical and spatial information in the images without, however, having to examine the entire images. To improve the querying efficiency, we propose to investigate in two directions: deriving higher-level indices through classification and providing a method that finds targets of interest through interactive learning. The first technique takes advantage of domain knowledge of underlying applications. The second technique accounts for quantification effect and noise in images and accommodates “learning from negative examples”, resulting into quite good discriminating power. Experimental results are given to demonstrate the effectiveness of our approach.
UR - http://www.scopus.com/inward/record.url?scp=84949977248&partnerID=8YFLogxK
U2 - 10.1007/3-540-60793-5_107
DO - 10.1007/3-540-60793-5_107
M3 - Conference contribution
AN - SCOPUS:84949977248
SN - 9783540607939
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 541
EP - 549
BT - Recent Developments in Computer Vision - 2nd Asian Conference on Computer Vision, ACCV 1995, Invited Session Papers
A2 - Li, Stan Z.
A2 - Mital, Dinesh P.
A2 - Teoh, Eam Khwang
A2 - Wan, Han
PB - Springer Verlag
T2 - 2nd Asian Conference on Computer Vision, ACCV 1995
Y2 - 5 December 1995 through 8 December 1995
ER -