Abstract | ||
---|---|---|
The need to quantify similarity between two groups of objects is prevalent throughout the signal processing world. Traditionally, measures such as the Kullback-Leibler divergence are employed, but these may require expensive computations of covariance or integrals. Maximum mean discrepancy is a modern distance measure that is computationally simpler - involving the inner product between the difference in means of two groups' feature distributions - yet statistically powerful, because these distributions are mapped into a high-dimensional, nonlinear feature space using kernels, whereupon the means are estimated via the Parzen estimator. We apply this metric and leverage several powerful data representations from the supervised image classification world, such as bag-of-visual-words and sparse combinations of SIFT descriptors, to locate scene change points in videos with promising results. |
Year | DOI | Venue |
---|---|---|
2013 | 10.1007/978-3-642-39094-4_38 | IMAGE ANALYSIS AND RECOGNITION |
Keywords | Field | DocType |
visual changepoint detection, unsupervised learning, maximum mean discrepancy, scene boundary detection, video indexing | Scale-invariant feature transform,Signal processing,Feature vector,Pattern recognition,Computer science,Unsupervised learning,Artificial intelligence,Contextual image classification,Estimator,Covariance,Computation | Conference |
Volume | ISSN | Citations |
7950 | 0302-9743 | 4 |
PageRank | References | Authors |
0.44 | 19 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Michael Diu | 1 | 26 | 1.60 |
Mehrdad J. Gangeh | 2 | 82 | 9.73 |
Mohamed S. Kamel | 3 | 4523 | 282.55 |