On parallel image retrieval with dynamically extracted features

Image retrieval with dynamically extracted features compares user-defined regions of interest with all sections of the archived images. Image elements outside the selected region are ignored, thus objects can be found regardless of the specific environment. Several wavelet- and Gabor-based methods for template matching are proposed. The improved retrieval flexibility requires immense computational resources, which can be satisfied by utilisation of powerful parallel architectures. Therefore, a cluster-based architecture for efficient image retrieval is discussed in the second part of the article. Techniques for the partitioning of the image information, parallel execution of the queries, and strategies for workload balancing are explained by considering the parallel image database Cairo as an example. The quality and efficiency of the retrieval is examined by a number of experiments.