Content-Based Image Retrieval for Content Analysis

O video repositories are becoming more common as we enter the era of digital video. As the number, size, and accessibility of digital video repositories grows, so does the need for video storage and retrieval systems. Television news and entertainment operations in particular find themselves in need of solutions for storing and retrieving large amounts of digital video. The dilemma of what is often termed asset management is frequently addressed in publications targeted at professionals working in television news and in video production and postproduction (e.g., Cointreau, 1998; Gruszka, 1999; Roberts, 1998; Stewart & Millerburg, 1998). Driven in part by a rapidly growing market for video asset management products, computer and information scientists in the past decade have turned their attention to developing systems that aim to index, categorize, and summarize video. Indeed, Content-Based Image Retrieval (CBIR) is now an area of specialization for many researchers in computer and information science.1 Developers of CBIR systems increasingly aim to automate their systems. CBIR developers are working to minimize the need for human coding of video, noting that human indexing and annotation is often time-consuming, costly, and error prone (Brunelli, Mich, & Modena, 1999; Ferman & Tekalp, 1998; Guimaraes, Correia, Oliveira, & Martins, 1998). Useful reviews of recent efforts to automate video indexing can be found in Aslandogan and Yu (1999); Brunelli et al. (1999); Furht, Smoliar, and Zhang (1995); Gauch, Gauch, Bouix, and Zhu (1999); Perry et al. (1999); Rui, Huang, and Chang (1999); Yeo and Yeung (1997); and Yoshitake and Ichikawa (1999). Social scientists have long been interested in studying film and television content. Perhaps the highest profile content analyses are those that assess the amount and nature of violence in television entertainment programming (e.g., Gunter & Harrison, 1998; National Television Violence Study 3, 1998). Other recent studies have assessed portrayals of women in prime-time television (Signorielli & Bacue, 1999), portrayals of older adults in television advertising (Roy & Harwood, 1997), portrayals of alcohol and tobacco use in music videos (Durant et al., 1997), and the use of sound bites in television news coverage of presidential candidates (Lowry & Shidler, 1998).