Interactive Image Retrieval by Graph Spectral Methods

在这项研究中，我们提出了一个有效的检索方法，多媒体数据，如图像和视频，并开发了一个用户友好的多媒体检索交互式系统。我们首先开发了多媒体数据的聚类和可视化的图谱方法。在这些方法中，高维多媒体数据投影到低维空间的特征向量的线性变换，它们被划分成集群，并显示在屏幕上。除了这个数据可视化系统，我们已经开发了一个加速方法的多媒体数据的相似性搜索。提出了一种过滤方法，用于检索与用户查询相似的数据。在阐述这两种多媒体数据可视化和快速检索的基本方法的基础上，我们开发了一个交互式的图像和视频检索系统。在该系统中，检索到的数据通过VRML显示在布置在三维空间中的屏幕上。此显示在用户每次选择图像或视频之后递增地更新。该系统引导用户在数据库中有效地查找目标数据。该检索系统的有效性已被不同用户的一些实验所验证。

We have proposed, in this research, an effective retrieval method for multimedia data such as images and videos and have developed an interactive system for user-friendly multimedia retrieval. We first developed graph spectral methods for clustering and visualization of multimedia data. In these methods, high dimensional multimedia data are projected into low dimensional space by a linear transformation of feature vectors and they are partitioned into clusters and are displayed on a screen. In addition to this data visualization system, we have developed an acceleration method for similarity search of multimedia data. A filtering method was developed for retrieval of data similar to user's query. Incorporating these two basic methods for visualization and fast searching of multimedia data, we developed an interactive system of retrieval of images and videos. In this system, the retrieved data are displayed on a screen arranged in a three dimensional space by VRML. This display is updated incrementally after each selection of an image or video by users. Users are navigated by this system for effectively finding target data in the database. The effectiveness of this retrieval system has been verified by some experiments with various users.