Mobile Intelligence

21.4 VIDEO DATABASE MODELING

A key component in this proposed study is the video database modeling mechanism called HMMM, which is deployed to bridge the semantic gap between the low-level video features and high-level concepts, to represent video temporal characteristics for event pattern queries, and to incorporate user preferences via feedback and learning strategies.

21.4.1 Markov Model Mediator

In essence, HMMM is developed based on Markov model mediator (MMM), a well-established mathematical model proposed in our earlier work [27]. MMM is represented by a five-tuple , where is a set of media objects called states, denotes the feature set, indicates the affinity relationships among media objects, represents the low-level feature values of media objects, and π indicates the likelihood of a media object being selected as the query. Here, a media object may refer to an image, a salient object, a video shot, and so on, depending on the modeling perspective and the data source. In addition, ...