A Visual Database System For Computer Vision Research

计算机视觉的数据库系统必须能够以多种形式处理数据，从标量特征的值到光强度数组，到表示部分和暂时识别的形状和模型的网络的复杂链接数据结构。这项工作需要设计和部分实现一个基于分层关系数据结构的原型VISION数据库系统，该结构可以处理各种类型的数据。该设计还包括一个新颖的可视化用户界面，允许用户检查和理解正在处理的实体。在这项工作中，通过研究机器人和医学这两个不同的视觉应用领域中使用的各种结构和操作，开发了数据库系统的组织和访问功能。在原型系统中实现了原语访问功能和一些高层视觉操作。针对机器人视觉和医学图像处理研究项目中的视觉算法，对系统的灵活性和效率进行了评估。其结果将是一个通用的科学数据库系统，它可以通过图形和应用程序接口访问，并将满足计算机视觉系统的需求。

A database system for computer vision must be able to manipulate data in numerous forms, running a full spectrum from values of scalar features, through arrays of light intensities, to complex linked data structures representing networks of partially and tentatively recognized shapes and models. This work entails the design and partial implementation of a prototype vision database system based on a hierarchical, relational data structure that can handle a wide variety of types of data. The design also includes a novel visual user interface that allows the user to examine and comprehend the entities being processed. In this work, the organization and access functions for the database system are developed through a study of the kinds of structures and operations used in two different vision application domains: robotics and medicine. Both primitive access functions and some higher-level vision operations are implemented in the prototype system. The flexibility and efficiency of the system are evaluated for vision algorithms being developed in robot vision and medical image processing research projects. The result will be a general scientific database system that can be accessed via both graphical and application program interface and that will satisfy the needs of computer vision systems.