摘要： 当检索样例位于数据库之外时，传统基于线性流形学习的图像检索方法在反馈迭代后的检索精度提高较小。为此，提出一种基于相关反馈和流形结构重构的图像检索方法。反馈时计算被检索样例的最邻近点，将被检索样例重构入需要保留的结构图中，从而满足映射时需保证相似图像和被检索样例距离尽可能近的要求。实验结果表明，该方法在额外耗时较少的情况下，能有效提高检索精度。

关键词: 流形学习, 基于内容的图像检索, 相关反馈, 流形结构重构, 维数约减, 语义流形

Abstract: In the case that sample is out of database, retrieval precision of the existing image retrieval methods based on linear manifold learning has smaller increase after feedback. Aiming at this problem, this paper proposes an image retrieval method based on Relevance Feedback(RF) and manifold reconstruction. It reconstructs sample into structure graph which needs to be reserved by computing the nearest neighbor in relevance feedback, thus can meet the need of that the distance between similar images and sample is mapped as near as possible. Experimental result shows that the method can improve retrieval precision while merely increase milliseconds time.

Key words: manifold learning, Content-based Image Retrieval(CBIR), Relevance Feedback(RF), manifold structure reconstruction, dimension reduction, semantic manifold

中图分类号: 

• TP911.73