Clustering of Visual Information Using Spectral Methods

Data clustering emerged over the last "digital" years as a powerful tool in data organization and indexing. Applications that use clustering methods for extracting a meaning from data, range from information retrieval, image processing, pattern recognition to biology, economic sciences and many more. This book can be used to learn basic information about data clustering for readers who want to get familiar with clustering techniques. In addition book shows how specific clustering algorithms can be applied in real world applications, especially in image and video processing. The main topic of the book is our proposed spectral clustering algorithm which is based on the Markov Random walk interpretation of pair-wise similarities between two data points. The proposed algorithm makes no assumption on the number of clusters present in the data set, instead it recursively uncovers clusters in the data until all groups have been found. We demonstrated the algorithm on a number of interesting problems in fields of image and video segmentation, shot and scene detection and video summarisation.

Autorentext

Uros Damnjanovic was born in Belgrade, Serbia and graduated at Faculty of Electrical Engineering. In 2009, he obtained his Ph.D. at Queen Marry University of London. Currently he is working as a post-doc fellow at Science and Technology in Archaeology Research Centre of the Cyprus Institute.

Klappentext

Data clustering emerged over the last "digital" years as a powerful tool in data organization and indexing. Applications that use clustering methods for extracting a meaning from data, range from information retrieval, image processing, pattern recognition to biology, economic sciences and many more. This book can be used to learn basic information about data clustering for readers who want to get familiar with clustering techniques. In addition book shows how specific clustering algorithms can be applied in real world applications, especially in image and video processing. The main topic of the book is our proposed spectral clustering algorithm which is based on the Markov Random walk interpretation of pair-wise similarities between two data points. The proposed algorithm makes no assumption on the number of clusters present in the data set, instead it recursively uncovers clusters in the data until all groups have been found. We demonstrated the algorithm on a number of interesting problems in fields of image and video segmentation, shot and scene detection and video summarisation.