[工程科技]ListofFigures.pdf

List of Figures 1 The visual interface of VideoQ. The figure shows an example query to retrieve video shots of all high jump sequences in the database. Note that the keywords “track and field” are used to filter the returned results. The results are shown in Figure 2 . . . . 6 2 The results: The displayed retrieved shots which include four ex- cellent matches (the first, the third, the fourth and the fifth), indi- cate the importance of the motion attribute in video shot retrieval. 7 3 The VideoQ system where the queries are in the form of animated sketches. Both the animated sketch and the browsing modes sup- port search with conjunction with keywords. . . . . . . . . . . . 26 4 The feature classification tree. . . . . . . . . . . . . . . . . . . . 27 5 Region segmentation and tracking of frame n . . . . . . . . . . . . 28 6 Region projection and segmentation of frame n . . . . . . . . . . 29 7 Region segmentation on QCIF sequences, using feature fusion. The top rows show the original sequence while the corresponding bottom rows show the segmented regions. . . . . . . . . . . . . . 30 8 Generating the candidate video shot list for a single object query. The first column of features shows the features of the query, while the second column shows the features across all objects in the database. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 9 The set A is the set that is obtained from the search while the set B refers to the ground truth. Precision and recall are defined using these two sets.