Image Retrieval Using ESNs and Relevance Feedback

In order to overcome "semantic gap" between bottom features and high-level semantic in the image retrieval, this paper introduces the echo state network to strengthen the mapping between the high-level vision content and the bottom visual feature and designs a feedback category screening strategy. We extract the feature of the queried image and get the characteristic vector of the image, through introducing the echo state network for image and constructing sample testing model for each kind of image data, we can calculate the category probability of the queried image, so as to achieve similarity discrimination of the queried image and library image. After users make feedbacks to the retrieval results and systems use related feedback algorithm to amend image feature, we retrieve and screen the category of the returned image. Experiments show that our retrieval architecture can achieve very good retrieval results.