Indexing, Object Segmentation, and Event Detection in News and Sports Videos - Multimedia Database Retrieval: Technology and Applications - page 207

via a multiple-level database access technique, which is highly desirable for news-
on-demand applications.
To enhance the robustness of video object segmentation, histograms of oriented
gradients may be incorporated into the conventional Graph Cut algorithm. This leads
to a new way of enhancing video segmentation accuracy, thereby incorporating
a shape prior into the algorithm for segmentation of pre-trained objects such as
humans.
In addition, the chapter presents an effective and robust method for detecting
faces in video sequences based on a coarse-to-fine strategy. A local normalization
technique is incorporated into a conventional face detector to alleviate the illu-
mination variation problem. It is demonstrated that the method can improve the
face detection rate and reduce the processing time. Compared with face detection
without local normalization, the current method has the following advantages: it
alleviates illumination variation problems, decreases computing time, locates faces
automatically on single frames to eliminate the manual initiation step, and is able to
detect a variety of faces reliably.
Finally, the indexing of NFL video using MPEG-7 descriptors is presented. The
current work develops a system with two main components. The first component
finds the starting points on play events within a video shot. The second component
is responsible for indexing and classifying events in American football. Both the
components of the system utilize MPEG-7 motion descriptors, while MPEG-7 audio
and MFCC features are added to enhance the indexing capabilities of the system.