基于多示例学习的图像检索方法

多示例学习对处理各类歧义问题有较好的效果，将它应用于周像检索问题，提出了一种新的基于多示例学习的图像检索方法。首先提取每幅图像的局部区域特征，通过对这些特征聚类求得一组基向量，并利用它们对每个局部特征向量进行编码，接着使用均值漂移聚类算法对图像进行分割，根据局部特征点位置所对应的分割块划分特征编码到相应的子集，最后将每组编码子集聚合成一个向量，这样每幅图像对应一个多示例包。根据用户选择的图像生成正包和反包，采用多示例学习算法进行学习，取得了较为满意的结果。

Multi-instance learning based approach to image retrieval

Multi-instance learning has better learning effects for all kinds of ambiguity problem, soapplying it to the image retrieval problem. This paper proposes a new method of image retrieval based onmulti-instance learning. First, it extracts local area features from each image, getting the basis vectors byclustering these features, and then utilizing them to encode every local feature. After that using the meanshift clustering algorithm for image segmentation, then partitioning each feature encodes into acorresponding subset according to the segmentation that local area feature points belong to. Last,aggregating each coding subset into a vector, so that every image is corresponding to a bag of instances.Next, query images posed by the user are transformed into corresponding positive and negative bags and amulti-instance learning algorithm is employed for image retrieval that achieves satisfactory results.