Shot Change Detection Based on Local Linear
Prediction in Time Domain
Cai Cheng1£¬Wang Ping1£¬Lam Kin Man2£¬Tan Zheng1
£¨1.School of Electronics and Information Engineering£¬ Xi'an Jiaotong University£¬
Xi'an 710049£¬China£»
2.Centre for Multimedia Signal Processing£¬ Department of Electronic and Information
Engineering£¬The Hong Kong Polytechnic University£¬ Hong Kong£¬ China£©
Abstract£ºA novel approach based on local linear prediction in time domain was proposed
to detect shot changes such as cuts and dissolves accurately and efficiently in a unified
framework. From the theoretical analysis and linear regression of video data£¬ two linear
prediction models of dissolves and stationary scenes were developed respectively. By
comparing the errors of these two models£¬ cuts and dissolves were detected
simultaneously. Since the analysis was carried out directly in the video compressed domain£¬
lots of decode time were saved. Besides£¬ the computational complexity was very low
because only the operation of addition£¬ subtraction£¬ comparison£¬ and shifting was
performed on integers. Therefore£¬ this detection is suitable for lowª²power devices and
embedded systemsª± Experimental results based on the MPEG-7 standard sequences
demonstrate that the recall and precision of cut detection are 95.5£¥ and 93.9£¥
respectively£¬ and the recall and precision of dissolve detection are 82.2£¥ and 76.0£¥
respectively. With this approach only 3.5 seconds were required for the shot change
detection of a 38ª²minute video.
Keywords£ºshot change£» cut detection£» dissolve detection£» linear prediction