1. Introduction
Blogging is a well recognized tool for communications and creation of a personal diary [1]. Consisting of a series of entries that describe or comment upon events, blogs could improve personal productivity by providing an easily browsed and searched chronicle of what individuals do during the work day. For example, a blog entry might describe a meeting with someone about a new invention. Access to those comments could be particularly valuable months or years later when refining the innovation or trying to recall precisely who invented what.
The effectiveness of a blog is significantly improved by the inclusion of multimedia – something that's becoming routinely available to office workers. Meeting recorders provide audio and video chronicles of office discussions [2]. Digital cameras could be used to take pictures of documents, whiteboards, business cards, etc. Key frames from a PC's desktop video could represent what a user does during the day. If office workers could easily access that multimedia data, it would be easy for them to maintain a multimedia diary.
Another issue for many people is the cognitive overhead of deciding that an event should be recorded with a multimedia capture device. By the time they realize that something should be chronicled, it's already occurred. This is a continuing source of frustration for users and is particularly significant in an office environment where it's important to record actions that can occur spontaneously, such as brainstorming discussions.
A significant technical issue in the creation of multimedia blogs is the automatic classification of images. Information about their content (for example, an image contains a whiteboard), would allow them to be automatically filed in a collection of whiteboard images. It would also simplify browsing since users often know the type of data they're looking for (it's a whiteboard but not a document) but don't recall the date it was captured.
We propose a novel solution for creating multimedia blogs that provides convenient access to many data sources and overcomes the cognitive burden of deciding when to capture an event with an "always-on" strategy. Our system, called the Office Blogger, or OBlog, includes a video camera for capturing office discussions, audio recording capability, automatic key frame extraction from PC desktop video, and a software interface to a digital camera that automatically classifies its images as whiteboards, documents, business cards, slides, and scenes. Multimedia is also automatically cross-linked with time and content in order to make retrieval of the information easier. Finally, a simple web interface enables users to easily browse and incorporate multimedia into their work blogs, creating multimedia-rich documents. Packaged as a stand-alone PC, this system adds no computational burden to the user's computer.
Previous work related to our system includes the MyLifeBits project, which is a repository for multimedia about a user's day-today life. Recent work on this project incorporates authoring tools for constructing stories [3]. In contrast, OBlog is designed for instantaneous capture and incorporation of multimedia in a diary where each entry is an independent multimedia comment about an event.
In this paper, we first give an overview of the OBlog system which allows classification and linking of user data from various sources. We describe the capture and indexing components of the system, i.e., how office conversations, PC-screen images, and printed documents are captured, processed, and indexed. Section 3 presents a novel algorithm that classifies business images. A user interface for creating the multimedia office blogs is presented in Section 4. Finally, we discuss the outlook in Section 5.
