| Content based video retrieval is still in its early years. Recent approaches first detect concept occurrences instead of basing the search on the presence of low level features, such as color histograms, of the multimedia content. Essentially, this establishes a logical representation of the content on top of the binary data of the video stream. However, the detection process will never be perfect. The aim of my PhD thesis is to create a probabilistic ranking model for the logical representation of the video content which uses the probabilistic detection result expressing which representation is present in reality. The created framework will not only use concepts but also a model for the language the speakers used in the video. |
