Loading...

Review and Improve the Performance of Video Scene Detection Methods

Mohseni, Hossein | 2016

594 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48706 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Sharif Khani, Mohammad; Gholampour, Iman
  7. Abstract:
  8. Due to the developments in technology, digital equipments, networks, storage devices, a lot of videos have been produced so far. Due to the dayly high growth rate of videos and their data, organizing and managing of them is difficult for applications such as interactive television, digital libraries, video surveillance, etc.However, for these kinds of usage from videos, at first the video will need to decompose into a hierarchical structure such as shot, scene, subscenes, scene and story unit. In order to achieve these goals, after the determination of shot boundaries and the allocation of key frames for each shot, similarity between each two shots are measured using key frames. According to the definition of the similarity criteria, similar shots are placed in one group that is called scene. Similar scenes can be classified in a category called single story unit.In this study, it is aimed to explore an extensive review of existing methods in detecting video scenes. In the following, a new and optimal algorithm is proposed that uses a combination of several efficient methods to detect video scenes with a high accuracy.
    The performance of the proposed algorithm in various aspects have been evaluated by applying it over a series of videos. The obtained results shows that the average F-measure of the proposed algorithm in the best case is 98% and its average value is 91%. These results shows an average performance increase of 6% in comparison with the other two mentioned previous algorithms in the thesis
  9. Keywords:
  10. Video Processing ; Video Analysis ; Scene Detection ; Shot Boundary Detection ; Keyframes Extraction

 Digital Object List

 Bookmark

No TOC