Video Abrupt Transition Detection Based on K-L Transform

Shot boundary detection is the first step of the video analysis and retrieval. In this paper we propose a new technique of abrupt transition detection based on K-L transform according to the characteristics of human visual system. Firstly, the algorithm uses K-L transform of the RGB color space to produce image parameter modal that is consisted of eigenvectors. Secondly, the algorithm calculates the frame distance with the first and second principle axis's direction differences in the two adjacent frames and weighting coefficients. Thirdly the algorithm uses a changeable width sliding window in front of current frame to detect abrupt transition, and at the same time disposes the static shot. The experimental results show that the proposed algorithm outperforms the traditional HSV histogram method, executes in real time and is robust to illumination changes or camera motion occur.