Remote Sensing Image Retrieval Based On Image Rank Similarity And Query-adaptive Feature Fusion

With the amount of remote sensing images increasing quickly,it is one of the most challenging and urgent problems how to accurately and quickly retrieve images that users are interested in among a large number of remote sensing images.Content-based remote sensing image retrieval(CBRSIR)that is an important method to solve this problem,has become the focus of research in the field of remote sensing applications in recent years.This paper proposes a method of improved query-adaptive fusion(IQAF)firstly.The method can assign different weights to each retrieval feature of different query images and achieve late fusion.In order to improve the performance of the similarity measure and make the fusion of multiple features effective,this paper studies IQAF method further and proposes a method for remote sensing image retrieval.Query-adaptive feature fusion(QAFF)algorithm in this method can assign the weights for each feature quickly and adaptively by the shape behavior of the normalized score curves.The similarity score of this method is Image rank similarity(IRS).The IRS is calculated by the retrieval result lists of the query image and the database images,which are obtained by retrieved in the image collection according to Euclidean distance.In order to improve the accuracy of remote sensing image retrieval,the method regards some top images that get high fused scores and the query image as the query class.The final result is obtained according to the image to query class similarity(IQCS)which is the similarity between the query class and the database images.The main contributions of this paper are summarized as follows:(1)The image rank similarity IRS uses the ranking information of retrieval results to measure the similarity between images and normalizes the similarity scores,which helps the late fusion of multiple features well.(2)The multi-feature fusion method QAFF calculates on-the-fly the different weights of each feature for different queries by evaluating the normalized score curves.It is independent on the retrieval database,which makes it well suited to dynamic systems.(3)The image to query class similarity IQCS considers the image clustering information,and is measured by image-to-class rather than image-to-image.It explores the inherent information of the class.The experimental results on two public remote sensing datasets show that the algorithms are robust and improve the accuracy of remote sensing image retrieval.