Open Collections

Design and implementation of a content-based video retrieval system

University of British Columbia

As computer application enters a multimedia era, video data is becoming an information source. In this thesis, we investigate various video indexing and retrieval techniques and describe an implementation of a prototype video retrieval system that supports queries based on either visual or semantic content of video data. In our system the basic unit for video indexing and retrieval is the shot. Several shot boundary detection techniques have been implemented and evaluated. Each shot is represented by one or more key frames. The image feature vector of each key frame usually has very high dimensionality. So first we reduce the dimensionality of the feature space using the Principal Component Analysis (PCA) technique because most existing multi-dimensional indexing structure grows exponentially in size as the number of dimension increases. We've successfully reduced the dimension of the data sets obtained from our test video sequences from 256 to less than 10 while preserving more than 90 percent of the distance information between data points. Then we choose adaptive bucket k-d tree to index the PCA-transformed data points. For the semantic annotation of video data, we propose a predicate-based annotation which has several advantage over the existing key word based annotation, including expressiveness and the ability to support inference. We also introduce a knowledge base which contains some common facts and rules that can be shared by different videos. Some of the annotations are derivable from other annotations using the rules in the knowledge base, so they need not to be explicitly added. In this way we can save a human annotator's work as well as storage space.

Thesis/Dissertation

application/pdf

2009-03-27

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/6591

Computer Science

University of British Columbia

UBCV

DSpace