Adaptive Algorithms for Interactive Multimedia

by identifying semantically meaningful entities can be arduous because it is difficult to simulate human perception. However, by creating an algorithm to respond interactively to user preference, content-retrieval systems can become more efficient and easier to use. O ver the past decade, the amount of existing multimedia data has exploded because of the proliferation of low-cost devices designed for capturing and encoding it. But there has also been a corresponding increase in the awareness about the challenges associated with managing that data. 1 Emerging applications designed to organize, retrieve, and abstract multimedia data require efficient technologies to be truly useful. Such applications are typically designed to help users interact more effectively with their content by addressing issues associated specifically with retrieval and personalization. 2 For example, the Moving Picture Expert Group has developed the MPEG-7 standard—formally called the Multimedia Content Description Interface—to provide a rich set of standardized tools for describing multimedia content. 3 However, even with such standards, managing multimedia content presents many unique challenges that we should address to make modeling and describing multimedia semantic content easier. Humans perceive and characterize content using high-level concepts—such as action, romance, comedy, or emotional factors—that are not directly related to the content's technical attributes. A piece of music, for example, can evoke many different feelings in different people. These feelings might be perfectly valid ways to label and later retrieve the content. But without an interactive, adaptive system, it would be almost impossible to use human feelings as a way to store and retrieve data. An effective way to organize, retrieve, and navigate multimedia information would be to use these kinds of human perceptions to rank the data, navigate it, then update it according to the user's perceptions. One of the interactive learning techniques that might lead to this kind of technology is called relevance feedback, a strategy originally developed in traditional text-based information retrieval systems. In a relevance-feedback approach, the human is part of the multimedia-management process, which means that the user evaluates the results provided by the system; the system then adapts its performance according to the user's demands and preferences. Relevance feedback, apart from accounting for the human subjectivity in perceiving the content, eliminates the gap between high-level semantics and low-level features, which are often used for content description and modeling. 4 Recently, researchers have extended relevance feedback algorithms from text-based information retrieval to content-based image retrieval (CBIR) …