1 / 1

Context: MPEG Video Files (6h for SBD task) Rough Data Extraction: Motion compensation vectors

Q a) CUT GRAD CUT. Objective: Rough Indexing: Real Time Shot Boundary Detection Rough Datas – Rough Solution Fast & Intelligent Browsing TRECVid Shot Change Transition Effects Classification. Context: MPEG Video Files (6h for SBD task) Rough Data Extraction:

davis-king
Download Presentation

Context: MPEG Video Files (6h for SBD task) Rough Data Extraction: Motion compensation vectors

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Q a) CUTGRADCUT • Objective: • Rough Indexing: • Real Time Shot Boundary Detection • Rough Datas – Rough Solution • Fast & Intelligent Browsing • TRECVid • Shot Change Transition Effects Classification • Context: • MPEG Video Files (6h for SBD task) • Rough Data Extraction: • Motion compensation vectors • DC Coefficients (I-Frames or Prediction Error) • Intracoded/outliers MB Map Robust Camera Motion Estimation: 6 parameters affine model robust estimator DeltaQ b) Frame Dissimilarity Measures: P-Frames MC: Motion Continuity Measure Q: Number of Intracoded Macroblock ∆Q: Derivative of Q D: P-Frames Dissimilarity Measure (#) SHOT BOUNDARY DETECTION IN THE FRAMEWORK OF ROUGH INDEXING PARADIGM AUTHORS: L. Primaux {primaux@labri.fr}; J. Benois-Pineau {jenny.benois@labri.fr}; P. Krämer {kraemer@labri.fr}; J-P Domenger {domenger@labri.fr} LaBRI CNRS UMR 5800 / University Bordeaux1 http://www.labri.fr/Recherche/ImageSon/AIV/ I-Frames DC I-Frames matching takes into account Luminance and chrominance WMSE: I-Frames Dissimilarity Measure Weighted MSE Transition Effects Classification: The classification consists in considering as “CUT” all peaks which are immediately followed by their opposite, with the tolerance of one I-Frame in between. Moreover the peak must not be preceded by a high positive value. Otherwise peaks are considered as “GRAD”. The decision based on the derivative of the number of intra-coded macroblocks showed an increased performance of 20% in recall compared to absolute value Q in (#). Shot Boundary Detection: We assume a Gaussian Distribution of D, which is online updated as follows: Detection threshold λ (Kp set to1.8): here in order to let Gaussian more reactive, and α is set to 0,15 Results: The most equilibrated result is a recall of 70.2% and a precision of 63.4%. The best precision is 73% for a recall of 65% and the best recall is 74% for a precision of 57%. run-sec decod-sec segm-sec  processor-type-and-speed 6732      3874      2858    Pentium 4 2.8GHz The index i is retained as shot border if: But still, to recognize D(i) as a shot change peak, we consider the previous shot change peak value D(j) and apply the following decision rule:

More Related