Digital Signal Processing Reference
In-Depth Information
can be used to improve the video object segmentation. More challenging tasks will
be involved due to the inconsistent motion. To achieve real time segmentation, peo-
ple usually make the trade-off between the segmentation quality and the efficiency,
which will directly reduce the performance of the object segmentation. Apart from
the problems raised in the image segmentation, fast and accuracy object tracking
and mask updating also need to address sufficiently.
The fourth major challenge in image/video segmentation is the need to develop
appropriate validation and evaluation approaches. The first task is the formation
of common databases where all algorithms can be compared with each other.
Fortunately, a good segmentation Dataset and Benchmark was recommended by
Martin [
64
], which contains a number of static images (i.e., 100 test and 200 train
images) with different objects and backgrounds. The second issue is the develop-
ment of evaluation technique. Most evaluation methods in the current literature are
based on the computation of the scores between the ground truth mask and the seg-
mented result. It is reasonable but not sufficient to address the segmentation quality.
Some researchers in the field have been trying to address this critical issues. One
effort that has been carried out in recent years is the work [
63
], which presented the
PR index to compare the obtained segmentation with multiple ground truth images
through soft nonuniform weighting of pixel pairs that accounts for scale variation in
human perception.
1.5
Summary
Image/video segmentation will have a major role in intelligent visual signal process-
ing in the decades to come. With the transformation of image analysis from human
interactive mode toward an unsupervised mode, segmentation is becoming an es-
sential tool for pattern recognition and computer vision. It will be very useful for
bridging the semantic gap between the low level feature and the semantic concepts.
Segmentation is a fundamental research topic that provides unique opportunities for
content based coding and media analysis. However, the challenges listed in the last
section should be adequately addressed so as to continue to be a key technology in
pattern recognition.
References
1. Ngan King N., Li H.: Semantic Object Segmentation. IEEE Communications Society Multi-
media Communications Technical Committee E-Letter, 4(6), 6-8 (2009)
2. Meier T., Ngan K. N.: Automatic segmentation of moving objects for video objects plane
generation. IEEE Trans. Circuits and Systems for Video Technology, 8(5), 525-538 (1998)
3. Li H., Ngan King N.: Automatic video segmentation and tracking for content-based applica-
tions, IEEE Communications Magazine, 45(1), 27-33 (2007)
