HKUST Institutional Repository >
Electronic and Computer Engineering >
ECE Master Theses >
Please use this identifier to cite or link to this item:
|Title: ||Key frame selection for video transcoding|
|Authors: ||Chau, Wing San|
|Issue Date: ||2005 |
|Abstract: ||Key Frame Selection is the process to select the most representative frames of a long video sequence based on a given set of rules. This research work presents a novel algorithm for selecting a set of representative frames in MPEG compressed video. We use luminance DCT DC value, macroblock type and motion vector information to determine the motion activity and visual content change in MPEG compressed video. The advantages of this algorithm are highly accurate video content description and efficient extraction of the required information in the MPEG compressed domain by low complexity inverse quantization and variable length decoding. A visual content based key frame selection is proposed. One of the advantages of this approach is that it does not depend on any predefined threshold or parameter. It gives a set of key frame based on an objective model of visual content flow. Another advantage is that the predefined number of key frame can be obtained by using the proposed algorithm.
This research work also presents a novel metric and a reliable algorithm for scene change detection. The new frame difference metric measures motion and content change across the video by using the motion vector and macroblock information in the MPEG compressed domain. With this new metric, a reliable scene change detection algorithm is discussed. Experimental results reveal that our scene change detection algorithm can successfully reduce the false alarms caused by local frame motion activity and at the same time increase the recall by amplifying the acceptance probability of the reliable scene change candidate.|
|Description: ||Thesis (M.Phil.)--Hong Kong University of Science and Technology, 2005|
xiv, 84 leaves : ill. ; 30 cm
HKUST Call Number: Thesis ELEC 2005 Chau
|Appears in Collections:||ECE Master Theses|
Files in This Item:
All items in this Repository are protected by copyright, with all rights reserved.