| 摘要 |
<p>A system for detecting semantically significant events in a video sequence. The system may include four primary portions, namely, a feature extractor (40), a probabilistic model using Hidden Markov Model (or other model) to integrate the features and determine boundaries (42), a commercial/non-commercial filter distinguishing commercial and non-commercial slow motion replay segments (44), and a summary generator for generating summaries based on the detected slow motion replay segments (46). The feature extractor (40) is preferably designed to extract a feature from color histograms at block (50), and three features from pixel-based differences at block (52). The pixel based technique allows flexibility in the content suitable for analysis. The features extracted at block (52) characterize the slow motion, still fields, and/or normal-speed replay components of a replay segment. The preferred technique used by block (52) is a pixel-wise difference operation on two subsequent fields. The feature extracted at block (50) characterizes the edit effect component, and is preferably determined based upon a RGB color histogram of each field.</p> |
