CISE Postdoctoral Program: Postdoctoral Research Associate in Experimental Science: Applying Computer Vision Methods to Image Databases

9503994 Samet一个新兴的研究领域是包含非传统数据的数据库与传统数据库的集成。为了有效地存储和检索非传统数据，需要新的索引方法。这些方法必须捕捉数据的性质，并根据这些信息进行排序。在本研究中，将符号图像整合到关系数据库系统的方法将扩展到更一般的图像情况。问题是如何将存储的图像与作为查询呈现给数据库的观察到的图像进行匹配，但该图像是从不同的视点获取的。利用计算机视觉中使用的方法来解决这些问题，并将这些结果集成到关系（或任何其他）数据库系统的解决方案将被研究。该解决方案基于几何不变量的使用。另一个研究目标是将这项工作扩展到其他非传统数据类型，如视频数据。随着视频正在成为信息系统中常见的数据类型，数据管理系统应该支持这种新的数据类型。重要的问题是如何从视频序列中提取上下文、空间和时间信息，以便根据视频数据的内容对其进行索引。本文将概述一种解决方案，该解决方案基于收集计算机视觉中用于此目的的方法的信息，并对其进行调整，使其可以在数据库系统框架中使用。特别建议使用从人脸视频序列中识别面部表情的系统作为测试平台。＊＊＊

9503994 Samet One of the emerging areas of research is the integration of databases containing nontraditional data with conventional databases. New indexing methods are required in order to store and retrieve nontraditional data efficiently. These methods must capture the nature of the data and sort on this information. In this research the methods for integrating symbolic images into relational database systems will be extended to a more general case of images. The problem is how to match the stored image with an observed image that is presented to the database as a query, but is taken from a different viewpoint. A solution based on utilizing methods that are used in computer vision to solve these problems, and integrating these results into a relational (or any other) database system will be investigated. The solution is based on the use of geometric invariants. An additional research goal that extends this work into other nontraditional data types such as video data will also be performed. As video is becoming a common data type in information systems, data management systems should support this new kind of data. The important issue is how to extract contextual, spatial and temporal information from the video sequences in order to index on video data based on its contents. A solution that is based on gathering information about methods that are used for this purpose in computer vision, and adapting them so they can be used in the framework of database systems will be outlined. In particular, using a system the recognizes facial expressions from video sequences of a human face is suggested as a testbed. ***