The design of a multi-concept image retrieval system based on Hadoop and GMM

The multi-concept image annotation and retrieval are popular for semantic image retrieval which needs to establish the relations between multi-linguistic labels and one image. So the probabilistic formulation for semantic labeling is introduced to solve it. However, the training process of the Gaussian Mixture Model classifiers needs a large amount of computation, especially when the image sets is huge. Hadoop is the open source software that has powerful parallel computing ability and big data processing capacity. In this paper, we introduced the GMM which is used for the likelihood computation for the linguistic indexing, and the Hadoop that can solve the computation complexity with the MapReduce framework. We will show our ideas, designs and realizations of this multi-concept image retrieval system.