|
The explosive growth of web videos prompts an urgent demand on efficient grasping the major events. Unfortunately, the unique characteristics of web video scenario, such as the limited number of features, the unavoidable errors of near-duplicate keyframe detection, the noisy text information, make web video event mining a challenging task. In this paper, we first explore the properties of textual feature trajectory from title/tags and visual feature trajectory induced from near-duplicate keyframes. Based on the study, we propose web video event mining solution by fusion of textual and visual feature trajectories, which takes into account peak time difference, overlap time span, and trajectory distance. Experiments on a large number of web videos from YouTube demonstrate the proposed method achieves good performance for web video event mining. |
|
Keywords:Applied Computer Technology; Feature trajectory; near-duplicate keyframes; event mining; web videos |
|