Clustering-Based Feature Selection for Content Based Remote Sensing Image Retrieval

摘要: During the last three decades, imaging satellite sensors have acquired huge quantities of remote sensing data. Content-based image retrieval is one effective and efficient techniques for utilizing those Earth observation data sources. In this paper, a novel approach, which based on feature selection semi-supervised learning, proposed. The new method includes four steps. Firstly, clustering employed to select features number clusters determined automatically by using MDL criterion; Secondly, according an improved validity index, we optimal can describe objectives efficiently; Thirdly, weights selected are dynamically determined; finally, appropriate learning scheme adaptively thus conducted. Experimental results show that, proposed approach achieve comparable performance relevance feedback method, while ours need no human interaction.