摘要：An efficient algorithm for shot boundary detection is proposed in this study. In order to eliminate the disturbances caused by illumination and camera motion, Dual Tree Complex Wavelet Transform (DTCWT) is used in the process of raw feature generation. Then, the enhanced texture features were extracted from the six detailed subbands by calculate the mean and standard deviation of the high-frequency coefficients. Non-negative Matrix Factorization (NMF) is used for extracting structure feature from the low-frequency sub-bands image and dynamic threshold is used for shot boundaries detection. The experiment result conducted on TRECVID 2001 test data and other test videos shows that the proposed scheme can not only overcome illumination and motion effects efficiently, but also achieve a high detection speed.
