Mixed Propagation for Image Retrieval

Dozens of image features have been proposed in recent decades, which could measure the similarity of images and promote the improvement of performance in image retrieval. Different features focus on different views of the image, where two image quite distant with one feature may close with another. In this paper, we attempt to integrate different measures together to improve the image retrieval accuracy. Main point of which is fusing several given measures into one and then propagating on the resulted data space. As different features are involved in the similarity measurement, it would help to retrieval more similar image for the query from the dataset. Experiment results on several image dataset show the benefit of the proposed algorithm.