CRI: CRD: Public web-based photo-collections as a research testbed

数字成像技术的快速发展创造了大量的图像在互联网上被收集、分类、搜索和共享。这反过来又加速了基于内容的图像管理系统的研究。这样的系统必须依靠计算机视觉和图像处理来提取图像描述符、检测物体和人脸、图像相似性度量、分类算法等。虽然提出了许多方法，但由于无法获得用于测试的“真实”图像集和缺乏实用的评估工具，进展受到阻碍。目前，用于对象/场景检测、视觉搜索和数据库导航的工具都是在很小的（有些武断的）图像集合上进行评估的，如果有的话，使用的用户数量也非常少。本提案寻求资金，为计算机视觉和图像处理社区开发一个资源，包括一个基于网络的公共图片集。对公众用户来说，它将是一个最先进的在线系统，可以组织、搜索和分享个人照片。然而，它的真正目的将是为研究人员提供一个大型的“真实”图像数据库进行测试，并为大量无偏见的用户进行评估。通过利用基于web的图像服务的蓬勃发展的公众需求，该资源将以前所未有的规模提供开发和评估基于内容的图像管理系统的各个组件的能力。除了提供研究资源外，该系统还将为计算机科学和工程专业的学生提供独特的教育体验。为此，该系统将包括一个二级“在线实验室”，可从主要网站访问，并允许同意的公众用户尝试实验服务。这个“在线实验室”将使学生有能力实施和测试他们的项目——不仅在计算机视觉和图像处理领域，而且在点对点系统、分布式计算、编程语言、安全和隐私以及界面设计等领域。该项目的进度报告将在http://www.eecs.harvard.edu/~zickler上定期更新。

Rapid development of digital imaging technology has created a flood of images being collected, sorted, searched and shared over the Internet. This in turn has led to the acceleration of research in systems for content-based image management. Such systems necessarily rely on computer vision and image processing for the extraction of image descriptors, the detection of objects and faces, image similarity measures, classification algorithms, etc. While many methods have been proposed, progress is being stifled by insufficient access to `real' image collections for testing and a lack of practical tools for evaluation. At the moment, tools for object/scene detection, visual search, and database navigation are being evaluated on small (somewhat arbitrary) collections of images, and if at all, using a very small number of users.This proposal seeks funding to develop a resource for the computer vision and image processing communities consisting of a public web-based photo collection. To a public user, it will be presented as a state-of-the-art online system for organizing, searching and sharing personal photos. It's true purpose will be for researchers, however, to whom it will provide a large `real' database of images for testing, and a large pool of unbiased users for evaluation. By exploiting the blossoming public demand for web-based image services, this resource will provide the ability to develop and evaluate individual components of content-based image management systems at an unprecedented scale.In addition to providing this research resource, the proposed system will enable unique educational experiences for students in computer science and engineering. For this purpose, the system will include a secondary `online laboratory' that is accessible from the principal website and allows consenting public users to try experimental services. This `online laboratory' will give students the ability to implement and test their projects---not only in the areas of computer vision and image processing, but in areas such as peer-to-peer systems, distributed computing, programming languages, security and privacy, and interface design.Progress reports for this project will be regularly updated at http://www.eecs.harvard.edu/~zickler.