Digital Signal Processing Reference
In-Depth Information
Tabl e 6. 1
Segmentation
quality matrix
[1-highly
desirable, 4-least desirable]
Scenario
Real time
Offline
User Interactive
2
4
Non-User Interactive
1
3
same, the extra time dimension in video makes segmentation in video different from
that in images. As stated by Zivkovic et al. [
23
] the vital problem in video analysis
is segmenting foreground object from background and in general, extracting high
level semantics from video could be a task of interest.
Ideally, any video segmentation application should address and satisfy the fol-
lowing two different essential properties:
1. Precision of object contours: This refers to how well the object boundaries are
correctly identified.
2. Temporal coherency of the partition: This refers to ability of segmentation algo-
rithm to identify the segment throughout the time to enable tracking.
Sometimes, very precise contours and high temporal coherency is required. Other
times, a rough identification of the object locations (e.g., using bounding boxes) is
enough. The key issue is to have accurate contour with consistent partition along
time. While segmentation quality is ultimate key factor to rate the performance
of any method, time and need for user interaction are important concerns to be
considered for segmentation. In [
9
], Izquierdo et al. has explained key components
of segmentation system.
Correia et al. has classified applications of video segmentation into set of scenar-
ios according to application constraints and goals [
3
]:
1. Real-time Nonuser Interactive scenario
2. Real-time User Interactive scenario
3. Offline User Interactive scenario
4. Offline Nonuser Interactive scenario
Li et al. [
13
] has classified video application into six categories such as
Video Surveillance, Content-based Video Summarization, Content-based Coding,
Computer
Vision,
Videoconferencing/Videophone
applications,
and
Digital
Entertainment.
As one can see from Table
6.1
, in addition to preserving the accuracy, any highly
desirable segmentation system should be (a) fully automatic, (b) able to work in real
time. The score in Table
6.1
reflects the same.
6.2
Recent Trends in Video Segmentation
Recently, there has been a strong surge for video segmentation in mobile media
applications such as augmented reality and mobile media communications such as
object based coding. A full classification of segmentation applications into a set of
