Evaluation of a M-Tree in a Content-Based Image Retrieval System

After validating an image feature extraction system (an Active Net) and created a metric for the extracted features (casting the Active Nets to graphs), it is necessary to make an index structure that speeds up the search process [1, 2], reducing the necessary I/O to disk. The M-Tree, due to its properties was selected as an indexing structure. The high dimensionality of the extracted features and its non-vector structure made the M-Tree one of the logical choices. The high dimensionality of the graphs and the non multidimensional vector structure of them made the M-Tree the logical choice [3].