ImageRank: A novel sorting algorithm with relevance feedback in application of national costume image retrieval
Baiyou Ma, Tianwei Xu, Juxiang Zhou · 2017
It is difficult to obtain fairly satisfactory result for national costume images retrieval in real application using existing search engine due to the complexity of visual features and semantic information implied in national costume. Given the fact, a novel sorting algorithm ImageRank is proposed for image retrieval with relevance feedback in this paper, inspired by the PageRank algorithm used by Google Search to rank websites or webpages in their search engine results. ImageRank is realized based on Lucene Image Retrieval (LIRE) toolkit, which is a Java library that provides a simple way to retrieve images and photos based on color and texture characteristics. And it assimilates a good deal of PageRank to accomplish efficient image retrieval by ensuring accurate level of images correlation in resource repository through update the indexical (points-to) relationship timely according to user's behavior dynamically. Experimental results investigate that the performance of ImageRank is better than LIRE. In addition, the conservativeness and convergence of IR value in ImageRank are also verified.