Medical Image Retrieval by Graph-based Semi-supervised Learning

As low level features can not reflect the high level semantic in medical image search, an image retrieval algorithm is proposed by graph-based semi-supervised learning frame.Firstly, agraph model is constructed by distance between images, and density similarity constrained in the label propagation progress is added to get the membership degree of query images, called visual semantic representation;then the dense SIFT feature of the image blocks is extracted and described with bag of visual words, in order to get the local feature;Finally, a combination of visual concept and local feature strategy is designed for similarity measurement.Experimental results of ImageCLEFmed database demonstrate that the proposed algorithm represents the visual semantic of images effectively, and achieves a better retrieval performance than single low level feature.