Share Email Print

Proceedings Paper

Video shot boundary detection using region-growing-based watershed method
Author(s): Jinsong Wang; Nilesh Patel; William Grosky
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

In this paper, a novel shot boundary detection approach is presented, based on the popular region growing segmentation method - Watershed segmentation. In image processing, gray-scale pictures could be considered as topographic reliefs, in which the numerical value of each pixel of a given image represents the elevation at that point. Watershed method segments images by filling up basins with water starting at local minima, and at points where water coming from different basins meet, dams are built. In our method, each frame in the video sequences is first transformed from the feature space into the topographic space based on a density function. Low-level features are extracted from frame to frame. Each frame is then treated as a point in the feature space. The density of each point is defined as the sum of the influence functions of all neighboring data points. The height function that is originally used in Watershed segmentation is then replaced by inverting the density at the point. Thus, all the highest density values are transformed into local minima. Subsequently, Watershed segmentation is performed in the topographic space. The intuitive idea under our method is that frames within a shot are highly agglomerative in the feature space and have higher possibilities to be merged together, while those frames between shots representing the shot changes are not, hence they have less density values and are less likely to be clustered by carefully extracting the markers and choosing the stopping criterion.

Paper Details

Date Published: 25 October 2004
PDF: 11 pages
Proc. SPIE 5600, Multimedia Systems and Applications VII, (25 October 2004); doi: 10.1117/12.570842
Show Author Affiliations
Jinsong Wang, Wayne State Univ. (United States)
Nilesh Patel, Univ. of Michigan/Dearborn (United States)
William Grosky, Univ. of Michigan/Dearborn (United States)

Published in SPIE Proceedings Vol. 5600:
Multimedia Systems and Applications VII
Chang Wen Chen; C.-C. Jay Kuo; Anthony Vetro, Editor(s)

© SPIE. Terms of Use
Back to Top