Pattern Vectors from Algebraic Graph Theory
IEEE Transactions on Pattern Analysis and Machine Intelligence
Dominant Sets and Pairwise Clustering
IEEE Transactions on Pattern Analysis and Machine Intelligence
Discovering Shape Classes using Tree Edit-Distance and Pairwise Clustering
International Journal of Computer Vision
Graph embedding using tree edit-union
Pattern Recognition
Clustering and Embedding Using Commute Times
IEEE Transactions on Pattern Analysis and Machine Intelligence
Fast communication: Dominant sets clustering for image retrieval
Signal Processing
Hierarchical Pairwise Segmentation Using Dominant Sets and Anisotropic Diffusion Kernels
EMMCVPR '09 Proceedings of the 7th International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition
Graph-based quadratic optimization: A fast evolutionary approach
Computer Vision and Image Understanding
Dominant sets based movie scene detection
Signal Processing
Evaluating minimum spanning tree based segmentation algorithms
CAIP'05 Proceedings of the 11th international conference on Computer Analysis of Images and Patterns
A robust graph partition method from the path-weighted adjacency matrix
GbRPR'05 Proceedings of the 5th IAPR international conference on Graph-Based Representations in Pattern Recognition
Probabilistic subgraph matching based on convex relaxation
EMMCVPR'05 Proceedings of the 5th international conference on Energy Minimization Methods in Computer Vision and Pattern Recognition
Hypergraph based information-theoretic feature selection
Pattern Recognition Letters
Unsupervised clustering of human pose using spectral embedding
SSPR'12/SPR'12 Proceedings of the 2012 Joint IAPR international conference on Structural, Syntactic, and Statistical Pattern Recognition
| Hi-index | 0.00 |
Dominant sets are a new graph-theoretic concept that has proven tobe relevant in partitional (flat) clustering as well as imagesegmentation problems. However, in many computer visionapplications, such as the organization of an image database, it isimportant to provide the data to be clustered with a hierarchicalorganization, and it is not clear how to do this within thedominant set framework. In this paper we address precisely thisproblem, and present a simple and elegant solution to it. To thisend, we consider a family of (continuous) quadratic programs whichcontain a parameterized regularization term that controls theglobal shape of the energy landscape. When the regularizationparameter is zero the local solutions are known to be in one-to-onecorrespondence with dominant sets, but when it is positive aninteresting picture emerges. We determine bounds for theregularization parameter that allow us to exclude from the set oflocal solutions those inducing clusters of size smaller than aprescribed threshold. This suggests a new (divisive) hierarchicalapproach to clustering, which is based on the idea of properlyvarying the regularization parameter during the clustering process.Straight forward dynamics from evolutionary game theory are used tolocate the solutions of the quadratic programs at each level of thehierarchy. We apply the proposed framework to the problem oforganizing a shape database. Experiments with three differentsimilarity matrices (and databases) reported in the literature havebeen conducted, and the results confirm the effectiveness of ourapproach.