Similarity Matrix

http://www.fxpal.com/systems/MediaAnalysis/sim10000.gif

This similarity matrix illustrates photo similarity in a study related to organization of digital photo collections.

The authors of the paper "Temporal Even Clustering for Digital Photo Collections" were attempting to address the growing use of digital photography and develop methods for developing user friendly software for organization and retrieval of photos. This matrix is from their paper. (Matthew Cooper, Jonathan Foote, Andreas Girgensohn, and Lynn Wilcox). http://www.fxpal.com/publications/FXPAL-PR-03-215.pdf

"The similarity matrix above visualizes the temporal similarity of a collection of 512 photos. To compute the matrix, K = 10000. For each matrix, a photo-indexed novelty score is computed by correlating a checkerboard kernel along the main diagonal of the similarity matrix as described in this presentation. Starting with the coarsest scale (largest K), peaks in this score are detected and a hierarchical set of event boundaries is generated. A confidence measure trading off intra-event similarity and inter-event dissimilarity is used to select a single level of the hierarchy for presentation to the user. The similarity matrix above visualizes the temporal similarity of a collection of 512 photos. To compute the matrix, K = 10000. " http://www.fxpal.com/?p=eventDetector