Poster Submissions


New Visual Information Classification and Clustering Approaches

Tomas Piatrik, Uros Damnjanovic and Ebroul Izquierdo Queen Mary University of London
Multimedia and Vision research Group

A fundamental step towards content based image and video annotation and retrieval is clustering and classification. Although there have been several proposals to cope with the underlying classification task, little research has been done on the use of bio-inspired algorithms and spectral methods for the classification of visual data. Organization of visual data is cornerstone for efficient browsing and retrieval of visual information which is necessary prerequisite for saving useful information contained in the visual data. We investigate the application of clustering algorithms for image classification and video summarization problem. Clustering approaches based on the behavior of real ants and on the algebraic properties of graph spectra are used for the grouping of data. The Ant Colony Optimization (ACO) learning mechanism is integrated into a COP-K-means approach to solve image classification problems. In the second proposal, we model the ability of ants to build live structures with their bodies in order to discover, in a distributed and unsupervised way, a tree-structured organization and summarization of the video data. Spectral methods focus on eigenvalues and eigenvectors of specially created similarity matrix to determine interesting properties of an underlying dataset. These properties are then used for detecting shot and scene boundaries within a video, and to build a summary that is presented to the user.

Hosted by the Knowledge Media Institute