An Improved Remote Sensing Image Retrieval MethodBased on Bag of Word Framework

Objective The Bag-of-Words(BoW)approach has lower retrieval accuracy performance for remotesensing image retrieval.The rate of correct results is very low especially when target textures are sim-ilar.To overcome this shortcoming,an indirect comparison between a query and target image basedon local invariant feature of the background image is proposed.The solution has two phases:theprocess of building a database and a retrieval process.After building a database,a search of the nea-rest neighbor feature in a query image in the feature space of the image dataset through an approximateKD-Tree is executed,then their relationship is recorded.The relationship can be used in calculatingsimilarity between a query and target image during a retrieval procedure.Experiments show that theproposed method has a better recognition performance than that of the BoW and,at the same time,needs less storage space for saving local invariant features.