Large-Scale Multimedia Retrieval and Mining [Guest editors' introduction]

Rahul Sukthankar Intel Labs and Carnegie Mellon University R ecent years have witnessed an explosive growth of multimedia data due to higher processor speeds, faster networks, wider availability of high-capacity mass-storage devices, and the advent of cloud computing. Stimulated by current work in scalable machine learning, feature indexing and multimodal analysis techniques, researchers are increasingly interested in exploring challenges and new opportunities for developing scalable approaches for multimedia retrieval and mining. The enormous scale of multimedia data is reflected in the following statistics: approximately 120 million digital still cameras were sold in 2010; video already accounts for more than half of all Internet traffic, with YouTube attracting more than 2 billion views per day and 24 hours of video uploaded every minute. This explosion of the amount of data, number of users, and availability of new resources has led to greater expectations for multimedia retrieval and mining in terms of effectiveness and efficiency, for which existing analysis approaches and systems typically don’t suffice. Scalability to large data collections poses a particularly significant challenge for current multimedia processing methods. For instance, only about one-third of the papers appearing in ACM Multimedia 2008’s content track are applicable to this scale of data collections. Meanwhile, the research interest in processing large-scale image, audio, and video collections continues to grow rapidly, given the increasing availability of Web 2.0 websites, surveillance videos, and both personal and enterprise multimedia archives. We believe the tipping point for largescale multimedia analysis is quickly approaching. This special issue samples the state of the art in large-scale multimedia analysis techniques and explores how advanced multimedia analysis can be leveraged to address the challenges in large-scale data collections. In particular, from a total of 20 submissions, we selected five representative articles, that investigate large-scale multimedia analysis theory and systems across multiple application domains, such as Web event detection, landmark detection, image annotation, musical content mining, and cloud computing.