C.-H. Wang, C.-N. Lee, and C.-H. Hsieh (Taiwan)
Multiple features integration, shot boundary detection, video segmentation, and self-organizing map (SOM)
Shot boundary detection which deals with a single feature is difficult to provide both high precision rate and recall rate. To enhance precision rate and recall rate in shot boundary detection multiple feature are needed. This paper presents a new method that combines spatial global, spatial local and temporal correlation features. In addition, unsupervised self-organizing map neural network is employed to learn three types of transitions; namely, potential cut transition, potential gradual transition and no shot transition. Finally, the cut and gradual transitions are selected from potential cut transition and potential gradual transition by a novel shot identification scheme. Experimental results show that the proposed method yields good performance.
Important Links:
Go Back
