Indexing, Object Segmentation, and Event Detection in News and Sports Videos - Multimedia Database Retrieval: Technology and Applications - page 178

Table 7.1 Story retrieval results, obtained by two querying methods: shot-to-story (STS) and
group-to-story (GTS)

Retrieval result (recall, precision)

Query Story Shot-to-story Group-to-story

Query 1 S7, S22, S31 (0.33, 1), (0.67, 0.50), (1, 0.50) (0.33, 1), (0.67, 0.50), (1, 0.50)

Query 2 S5, S8, S14, S28, S29 (0.20, 1), (0.40, 0.67), (0.60,
0.50), (0.80, 0.11), (1, 0.10)

(0.20, 1), (0.40, 0.50), (0.60,
0.50), (0.80, 0.50), (1, 0.23)

Query 3 S1, S26, S46 (0.33, 1), (0.67, 1), (1, 0.75) (0.33, 1), (0.67, 1), (1, 1)

Query 4 S36, S37, S50, S55 (0.25, 1), (0.50, 1), (0.75, 1),

(1, 0.50)

(0.25, 1), (0.5, 1), (0.75, 0.38),
(1, 0.31)

Query 5 S7, S36, S39 (0.33, 1), (0.67, 0.40), (1, 0.50) (0.33, 1), (0.67, 0.52), (1, 0.43)

retrieved within the first ten retrievals. In Figure 7.7 , a comparison of STS and GTS
performances shows that STS performed slightly better than GTS at lower recall
levels, while GTS was much better at higher recall levels. This is opposite to the
results from retrieving video groups. In this case, a great deal of information from
the video group is favorable for retrieving the video story.

7.4 Segmentation of Video Objects

7.4.1 Graph Cut Video Segmentation

The automatic extraction of an object of interest in a video has immediate important
applications in video editing and indexing. The development of ways to segment a
video object accurately, efficiently and with minimal user interaction is an ongoing
research problem. One of the most prominent examples of the use of a video object
copy and paste is the blue screen background used by many television studios to
allow for simple color keying. This takes advantage of the fact that colors such as
blue and green are rare in an indoor setting like a news anchorman. This process is
simple and can be done in real-time; however it is limited by the use of a background
color. It requires the setup of a screen in a studio setting, and does not generalize
well to outdoors or other types of complex backgrounds.
Figure 7.8 shows a user program interface for segmentation of objects in news
videos. A user can get a rough estimate of a video object by just marking one frame.
The system allows the user to mark objects of interest in blue, and the background
in red. Then, the system automatically segments all following frames and shows
the result of an anchor object segmented. Similarly, Fig. 7.9 shows the result of
object segmentation, where the user marks the sailboat in Frame 2. The system
automatically tracks and cut the same object past Frame 30, even if the object moves
and changes its shape. In general, tracking and segmentation continues until a scene
changes or when there is severe occlusion.