content-based search to retrieve desired video clips. In the
first stage, a simple segmentation method is used to split
the archived digital video into five-minute video clips.
The system then extracts twenty image frames evenly
from each of these five-minute video clips for visual
feature extraction. Additionally, the system splits the
audio channel of each clip into twenty individual fifteensecond
segments for further audio feature extraction.
In the second stage, both audio and visual features are
extracted. For visual features, the color element is used
as the content feature. For audio features, 154 audio features
originally used by Ellis and Lee (2004) to describe
audio segments are computed. For each fifteen-second
video segment, the visual feature vector extracted from
the sample image and the audio feature vector extracted
from the corresponding audio segment are combined into
a single feature vector. In the information retrieval stage,
the user submits a video clip query at first, then its feature
vector is computed and compared with that of video clips
in the database, and the most similar clips are retrieved
content-based search to retrieve desired video clips. In thefirst stage, a simple segmentation method is used to splitthe archived digital video into five-minute video clips.The system then extracts twenty image frames evenlyfrom each of these five-minute video clips for visualfeature extraction. Additionally, the system splits theaudio channel of each clip into twenty individual fifteensecondsegments for further audio feature extraction.In the second stage, both audio and visual features areextracted. For visual features, the color element is usedas the content feature. For audio features, 154 audio featuresoriginally used by Ellis and Lee (2004) to describeaudio segments are computed. For each fifteen-secondvideo segment, the visual feature vector extracted fromthe sample image and the audio feature vector extractedfrom the corresponding audio segment are combined intoa single feature vector. In the information retrieval stage,the user submits a video clip query at first, then its featurevector is computed and compared with that of video clipsin the database, and the most similar clips are retrieved
การแปล กรุณารอสักครู่..
